-
2026-10-21 Strongly log-concave measures and lattices.
Summary: Let $Λ\subset \mathbb{R}^n$ be a full rank lattice with the property that $\det Γ\ge 1$ for all sublattices $Γ\subseteq Λ^*$. We prove that there exists a symmetric random vector $X \in \mathbb{R}^n$ with a smooth, strictly positive density $f : \mathbb{R}^n \to (0, \infty)$ satisfying that $f$ is $α$-strongly log-concave, where $α= 2π/e$, and $X + Λ$ is uniform on $\mathbb{R}^n/Λ$. Moreover, we show that the existence of such a density $f$ characterizes the assumption on $Γ$ up to a constant. There are a few consequences of the above. First, for such lattices we can bound the covering radius \[ μ(Λ, K) = \max_{x \in \mathbb{R}^n} \min_{λ\in Λ} \|x-λ\|_K \le \sqrt{\frac{2e}π} \cdot \mathbb{E}_G\, \|G\|_K \] of a convex body $K \subset \mathbb{R}^n$ containing $0$ in its interior, where $G \sim N(0,I)$. This implies an $O(n\sqrt{\log n})$ bound for the flatness constant and an $O(\log^{3/2} n)$ bound for the subspace flatness constant. Second, we establish as a consequence an alternative proof of the reverse Minkowski theorem with an explicit constant.
-
2026-10-20 Adversarial Benchmarks and BenchDoor Attacks: Exploiting Benchmark Incentives for Backdoor Attacks.
Summary: Machine learning benchmarks act as guideposts for model development and spur the creation of training datasets intended to improve benchmark scores. We introduce BenchDoor, a family of data poisoning attacks that exploits this incentive. Beyond preserving benign performance, BenchDoor requires benchmark gains that encourage adoption of data that also implants a backdoor. The adversary either manipulates a benchmark's training or test data, creating an adversarial benchmark, or supplies an independent training dataset while leaving the benchmark unchanged. We formalize five attack scenarios, including training on rewritten test examples and applying similarity-based decontamination. Across multiple-choice Q&A, text classification, and open-ended Q&A with Qwen and Gemma models, BenchDoor achieves positive benchmark gains for all 30 evaluated configurations, with held-out attack success rates of 82.33% to 100.00% at poisoning rates of 2% to 10%. Both objectives remain achievable after synthetic data generation and similarity-based decontamination, with attack success rates of at least 90.33%. These findings show how benchmarks can serve as attack vectors both through their own data and through training resources that deliver improvements on them.
-
2026-10-11 Instance-Adaptive Online Multiclass Multicalibration.
Summary: We study online multiclass multicalibration over $T$ rounds, and give an algorithm with beyond-worst-case guarantees. With $n$ possible outcomes, forecasts lie in a simplex of dimension $d=n-1$. In the worst case, cumulative $\ell_1$ multicalibration error scales with $\widetildeΘ(T^{(d+1)/(d+2)})$, which becomes untenable once $d$ exceeds a small constant. Our algorithm adapts to the geometry of the realized sequence: if the sequence of adversarially selected outcome means lies in an affine subspace of dimension $s$, it obtains error $\widetilde O(n^{1/(s+2)}T^{(s+1)/(s+2)})$ with high probability. The algorithm adapts to this structure without knowing the subspace or its dimension. For $T\ge n\ge2$, this rate is minimax optimal up to polylogarithmic factors. For a stochastic instance with a fixed conditional mean, the bound specializes to $\widetilde O(\sqrt{nT})$, giving a square-root dependence on the horizon.
-
2026-10-05 Finding Gaussian Structure in Bosonic States.
Summary: We study agnostic tomography of pure bosonic Gaussian states: given copies of an arbitrary $n$-mode bosonic state $ρ$, the goal is to output a pure Gaussian state whose infidelity with $ρ$ is at most $\mathrm{opt} + ε$, where $\mathrm{opt}$ is the minimum infidelity achievable by any pure Gaussian state. We give efficient protocols achieving this in both the high and low fidelity regimes. When $\mathrm{opt}$ is below some universal constant, our protocol has runtime and copy complexity which is strongly polynomial in $n, 1/ε$ and $\log \log E$, where $E$ is the energy of the closest pure Gaussian state. For arbitrary $\mathrm{opt}$, our protocol uses $(n+1)^{\mathrm{poly}(1/ε)} \mathrm{poly}\left(1+\log\log(E)\right)$ copies and runtime. As a corollary, we obtain the first truly tolerant Gaussianity testing protocol for distinguishing whether $\mathrm{opt} > c + ε$ or $\mathrm{opt} < c - ε$, for any threshold $c\in(0,1)$. We also prove $\mathrm{poly}(n,1/ε)$ runtime is impossible, unless $\mathrm{NP}\subseteq\mathrm{BQP}$. Our protocols follow a shared paradigm: first, we iteratively use general Gaussian measurements combined with techniques from classical robust statistics to obtain a good warm start estimate, then we leverage non-Gaussian measurements to refine this warm start using convex and non-convex optimization methods. Interestingly, we prove that non-Gaussian measurements are necessary to match the strong agnostic guarantees we obtain, and in fact these guarantees are provably superior to what is possible for robustly estimating classical Gaussians.
-
2026-10-05 Quantum 1-PCA with Pauli Measurements in Nearly Linear Time.
Summary: We consider the problem of quantum 1-PCA: given copies of an unknown $n$-qubit mixed state, recover a classical description of its leading eigenvector. Our goal is to do so using non-adaptive and single-qubit measurements. For an $n$-qubit state with top eigenvalue $λ$ and spectral gap at least $Δ> 0$, we give an algorithm that recovers the leading eigenvector to fidelity at least $1 - \varepsilon$ with high probability using $\tilde{O}\left( {2^n \cdot η^2} / {Δ^3 \varepsilon^3}\right)$ copies and $\tilde{O}((2^n/Δ\varepsilon) \cdot \operatorname{poly}(η/Δ\varepsilon))$ time, where $η= \max (1 - λ, \varepsilon)$. All of our measurements are non-adaptively chosen, and performed in single-qubit Pauli bases. When the spectral gap is constant and the desired accuracy is comparable to the noise level, i.e. $\varepsilon = Ω(η)$, our runtime and copy complexity become $\tilde{O} (2^n / \varepsilon)$. This generalizes the guarantees of Grewal et al. [arXiv:2601.04444], who achieved similar rates, but under the assumption $η= 0$, i.e., that the state was pure. Our results show that the same rates hold in the presence of state misspecification, up to polylogarithmic factors. From a technical perspective, our algorithm works by recursively constructing low-dimensional subspaces that approximately preserve the target eigenvector. To achieve nearly linear runtime dependence on the dimension of the Hilbert space, we develop a novel structured Pauli sampling scheme that enables fast batched computation of exponentially many projected Pauli matrices.
-
2026-10-05 Sample-Optimal Estimation of the Fréchet Inception Distance.
Summary: The Fréchet Inception Distance (FID) is widely used to evaluate generative models, but its empirical plug-in estimator suffers from finite-sample bias [BSAG18, CF20]. We study the sample complexity $n$ of estimating FID to error $ε$ between $d$-dimensional Gaussians with bounded mean distance and covariances, when one distribution is known. Our contributions are threefold. (1) We establish tight finite-sample $Θ(\frac{d^2}{n})$ bias and $Θ(\frac{d}{n} + \frac {d^2} {n^2})$ variance bounds for the empirical plug-in estimator, establishing a $\gtrsim d^2$ sample complexity. (2) To debias the empirical plug-in estimator, we generalize the ${\rm FID}_\infty$ estimator of [CF20] to extrapolation methods of arbitrary order $k$. We further prove tight bias and variance bounds of $Θ(\frac{d^{k + 2}}{n^{k + 1}})$ and $Θ(\frac d n + \frac{d^2}{n^2})$ for any order-$k$ extrapolation under our framework. (3) We introduce Relative Taylor Debiasing (RTD), a new, computationally efficient FID estimation algorithm using debiasing techniques inspired by U-statistics. We show that RTD achieves an $O(\frac d {ε^2})$ sample complexity, and prove that this is optimal. We provide a complementary empirical evaluation of our new estimators. Our experiments on synthetic Gaussians validate the predicted residual bias and support the tightness of our bounds. On ImageNet with Inception embeddings, RTD achieves the lowest mean estimation error at the standard 50K sample budget, while our second-order variance-aware extrapolation estimator (VALE$_2$) uses only 10K samples to achieve accuracy comparable to FID$_\infty$ at 50K samples.
-
2026-10-04 Gap Amplification for Local Hamiltonians with Combinatorial Soundness.
Summary: The quantum PCP conjecture is one of the major open problems in quantum complexity theory. It has resisted attack in part because many primitives used in the proof of the classical PCP theorem, such as locality-preserving gap amplification and alphabet reduction, have no obvious quantum analogues due to quantum no-cloning. Locality-preserving gap amplification is a procedure that takes as input a local Hamiltonian problem instance and produces a new instance with a larger promise gap, without increasing the locality of the Hamiltonian, and instead moderately increasing its local qudit dimension. Obtaining this kind of control over the locality during gap amplification is critical to the success of many known strategies for proving the classical PCP theorem. In this work, we put forth the first known viable template for quantum locality-preserving gap amplification, and we prove that our procedure amplifies the combinatorial gap of local Hamiltonians. Our work introduces a new framework for reasoning about quantum gap amplification in terms of fault-tolerant computation, and illuminates a route toward importing one of the central ingredients in classical PCPs into the quantum setting. In particular, we build upon ideas from the recent classical PCP of Bafna--Minzer--Vyas based on high-dimensional expanders, and the circuit-to-Hamiltonian construction of Anshu--Breuckmann--Nguyen, in addition to several recent advances in quantum coding theory.
-
2026-10-02 Three-Move Blind Signatures from DL.
Summary: This paper considers the problem of building blind signatures in pairing-free groups. We provide the first three-move blind signature which is provably one-more unforgeable, in the random-oracle model, under the minimal assumption that the discrete logarithm (DL) problem is hard. Our construction in fact achieves one-more strong unforgeability, and also supports partial blindness. Blindness is statistical, also in the ROM. Our construction makes black-box use of the underlying group and does not rely on non-black-box techniques. For most such constructions, three moves are necessary by a recent result of Dietz, Kastner, and Tessaro (CRYPTO '26). We build on recent work by Chairattana-Apirom, Reichle, and Tessaro (CRYPTO '26), which achieved the same combination of round complexity and security guarantees, but under the stronger decisional Diffie--Hellman (DDH) assumption. In particular, we follow the same paradigm of boosting the security of a weakly secure scheme (namely, the Okamoto--Schnorr blind signature) to a fully secure one by authenticating the initial nonce. We provide, however, a substantially different instantiation of this paradigm that dispenses with the use of DDH and establishes security from DL alone.
-
2026-10-01 Gacha Decoding: Eliciting Diverse Generations Through Instruction Following.
Summary: We introduce Gacha Decoding, an inference-time method for eliciting diverse language model generations that scales with model capability. Across open-ended domains (in-the-wild chat, creative writing, planning for image generation, and protein design), Gacha Decoding significantly outperforms existing generation diversity approaches at equal quality (up to 2.4x Vendi over the next-best prior approach), reaching the same number of high-quality modes with over an order of magnitude fewer samples (11.0x) and discovering novel modes that no other approach surfaces. Our key insight is to treat diversity as an instruction-following problem: rather than relying on the LM's token entropy, we combine its instruction-following capability with randomness from an external RNG tool to scalably identify and realize distinct modes of the response space. This approach of "planning with dice" enables Gacha to invert the long-observed tension between diversity and model capability. As the underlying LM becomes a better instruction follower, diversity under Gacha Decoding consistently improves--even as its token entropy and diversity under prior approaches decline. Together, our results highlight that instruction following, rather than token entropy alone, can drive generation diversity.
-
2026-09-30 All Unitaries Have Constant Depth Quantum Circuits.
Summary: It is well-known that every $n$-qubit unitary can be implemented by a $2^{O(n)}$-depth quantum circuit using single- and two-qubit gates. It has been open whether exponential depth is *necessary* for general unitaries, even when allowing an unlimited number of ancilla qubits. Here we show, perhaps surprisingly, that all unitaries can be implemented exactly by a circuit of one- and two-qubit gates of depth $\mathsf{poly}(n)$ with $2^{O(n)}$ ancilla qubits. In other words, every $n$-qubit unitary can be parallelized to polynomial depth. In fact, our depth bound is *linear* in $n$, which is the best possible, and an exponential improvement on the previous best bound of $2^{n/2}$ due to Rosenthal [TQC 2022, Quantum 2026]. Moreover, if we allow unbounded fan-out gates, these circuits can be further reduced to *constant* depth. Our construction takes advantage of a novel relationship connecting the unitary synthesis problem of Aaronson and Kuperberg to locally-decodable codes and private information retrieval from complexity theory and cryptography, and has a natural interpretation in bosonic quantum computation.
-
2026-09-30 A Noise Operator Approach to Quantum Query Complexity and Time-Space Tradeoff Lower Bounds.
Summary: Time and space (memory) are two of the most important measures of cost in computation, even more so for quantum computation. Yet, our tools for proving unconditional quantum tradeoffs between time and space are surprisingly limited. The first quantum time-space tradeoff lower bounds were proven for sorting by Klauck, Špalek and de Wolf. Unfortunately, their method is limited to proving output-oblivious lower bounds (i.e. the lower bounds only apply to algorithms with a non-adaptive output schedule) and other methods have yielded nothing beyond output-oblivious lower bounds for sorting. Here, we prove the first fully general quantum time-space tradeoff lower bound for sorting. We do so by introducing a novel method based on the noise operator to add to the analysis toolkit for proving quantum query and time-space tradeoff lower bounds. By combining our resulting quantum noise stability bound with quantum recording query methods, we prove an $Ω(n^{4/3} (\log \log n)/(S^{1/3} \log n))$ lower bound on the number of queries that a fully general quantum algorithm with at most $S$ qubits of memory requires to sort $n$ numbers from $[n^2]$. Applying our noise operator argument involves purely classical reasoning, which makes it particularly simple to use. We also use it to prove that, for any strongly universal (pairwise independent) hash function family $H$ from $n$ bits to $m$ bits, almost all hash functions in $H$ require any algorithm with at most $S$ qubits of memory to make $Ω(nm/S)$ quantum queries to input $x$ in order to compute $h(x)$, even with very small success probability. Previously, Mansour, Nisan, and Tiwari had shown a matching classical lower bound for computing $h(x)$ with both $h$ and $x$ as inputs using their hash mixing lemma. Our noise operator method allows us to use a related property of hash functions to prove our quantum lower bounds.
-
2026-09-30 A Cryptographic Perspective on Fingerprinting Machine Learning Model Weights.
Summary: We introduce a cryptographic framework for fingerprinting machine learning models, enabling model providers to train models that can later be attributed to them. Our framework captures an honest setting in which providers distribute base models that users may subsequently modify to adapt them to their own applications (e.g., through finetuning). We formalize three properties for fingerprinting schemes: (1) quality preservation or undetectability: fingerprinted models must remain computationally indistinguishable from models produced by an underlying training algorithm, (2) robustness: the fingerprint must remain detectable even after the model weights are subject to permitted modification, and (3) unforgeability: no computationally bounded trainer can forge the fingerprint except by applying a permitted modification to an existing fingerprinted model. Under the continuous learning with errors assumption (Bruna et al, 2021), we construct a robust and unforgeable fingerprinting scheme for modifications in $\ell_2$. The scheme achieves undetectability with respect to any given noisy training algorithm, meaning any algorithm whose output weights can be decomposed into independent Gaussian and non-Gaussian components. Along the way, we introduce two new technical tools. The first is a form of steganography; we develop a method for undetectably modifying a noisy training algorithm so that its output weights secretly encode a message. The embedded message remains decodable even after bounded $\ell_2$ perturbations to the weights. Second, we introduce the pancake alignment problem, a "semi-search" variant of homogeneous continuous learning with errors (hCLWE). Given hCLWE samples generated from a random secret, the goal is to recover a direction that is sufficiently aligned with the secret. We provide evidence for the hardness of pancake alignment in certain parameter regimes via reductions from continuous learning with errors.
-
2026-09-27 The Breakdown of Classical Minicrypt Equivalences in the Quantum-Computation Classical-Communication Model.
Summary: Classically, it is well-known that several fundamental cryptographic primitives, including one-way functions, pseudorandom number generators, commitments, and signatures, characterize the same cryptographic world, which is known as "Minicrypt". In this paper, we investigate to what extent this picture persists in the quantum-computation classical-communication (QCCC) setting, where parties may perform quantum local computation, but communicate through only classical messages. We demonstrate that the classical Minicrypt landscape breaks down in the QCCC model by oracle separations. We construct two oracle worlds where efficiently verifiable one-way puzzles (EV-OWPuzz) exist. By a known equivalence, QCCC one-time signatures also exist in both worlds. In the first, one-way functions do not exist, even if we allow quantum pseudodeterministic evaluation. In the second, QCCC bit commitments do not exist. Thus, the classical Minicrypt reductions from signatures to one-way functions and bit commitments have no fully black-box counterparts in the QCCC setting. Our proofs develop techniques for analyzing the random phase states, including a concentration theorem and an LOCC decoupling theorem. As a separate result, we establish quantum sum-binding for the NOVY protocol, settling the question of its post-quantum security. We also generalize it to a fully black-box construction of a QCCC commitment from EV-OWPuzz with nearly uniform puzzle distribution, complementing our oracle separation results.
-
2026-09-25 Hidden-Bits Generators with Succinct CRS and Openings.
Summary: Feige, Lapidot, and Shamir [FOCS'90] presented a generic approach for building non-interactive zero-knowledge proofs via the notion of hidden-bits generators (HBGs). At a high level, an HBG allows a prover to statistically commit to a $k$-bit pseudorandom string in the CRS model with a succinct commitment. Later, the prover can open this pseudorandom string at an arbitrary subset $S \subseteq [k]$ of positions by providing a proof $\pi_S$. The hiding property requires that unopened positions remain pseudorandom. In this work, we study the efficiency of HBGs, focusing on direct constructions that make black-box use of cryptography. We obtain the following results: Succinct CRS: Assuming learning with errors (LWE), we construct an HBG with {\em succinct} CRS of size $\mathsf{poly}(\lambda,\log k)$ for generating $k$ hidden bits. This achieves an exponential improvement over a recent sequence of works by Waters [STOC'24], Branco et al.~[EUROCRYPT'25], and Waters, Wee, Wu~[EUROCRYPT'25] that required a CRS of size at least $k \cdot \mathsf{poly}(\lambda)$. Batch opening: Using pairings, we construct an HBG with batch opening, where the proof size for any subset is $ \mathsf{poly}(\lambda)$. Previously, this was only known from the Naccache-Stern public-key cryptosystem [Groth, ASIACRYPT'10]. Succinct CRS and Batch opening: Assuming the hardness of RSA, we construct an HBG where both the CRS and the opening proof for any subset are of size $ \mathsf{poly}(\lambda,\log k)$. No black-box constructions achieving this efficiency were previously known.
-
2026-09-24 On FROST and Unconditional LDVR Security.
Summary: Inspired by the recent polynomial-time adaptive attacks on threshold Schnorr signatures, we contextualize the Low-Dimensional Vector Representation (LDVR) assumption and its impact on relevant threshold Schnorr signature schemes. Our analysis focuses in large part on FROST, a widely deployed threshold Schnorr signature scheme. However, our analysis is applicable to any relevant threshold Schnorr signature scheme, such as Lindell's three-round threshold Schnorr protocol. We show that for cases where FROST is used with 131 parties and fewer, and for parameter choices of 128-bit generic (discrete log) security, FROST is secure for any signing threshold. This bound increases to 260 parties for parameter choices of 256-bit security; our analysis similarly can be extended to other parameter choices. This bound assumes the best possible LDVR adversary, even more powerful than the results by Rodríguez García and Tessaro. In other words, for real-world uses of FROST with parameter choices of 128-bit security, and with a small (131 and fewer) number of parties, the security of FROST is determined entirely by the hardness of the Algebraic One-More Discrete Logarithm (AOMDL) assumption, as LDVR is unconditionally hard for this parameter regime. It is an open research question whether the concrete attacks given by Rodríguez García and Tessaro can be improved to approach the unconditional LDVR security bound considered in our analysis. We then show for each additional party for 132 total number of signers and greater, again for parameter choices of 128-bit generic security, the resulting security of FROST is at worst reduced by approximately one bit (or less) of security for the worst-case choice of corruption threshold. This analysis can be used by practitioners to determine the worst possible security of even large-scale FROST deployments. Finally, we give two variants to FROST that build upon prior masking techniques to allow for full adaptive security. The first variant allows for masking keys to be derived from Diffie-Hellman identity keys, allowing for constant memory overhead, at the tradeoff of increased computation. The second variant allows for a 50% reduction in memory overhead for storage of pairwise masking keys, by employing symmetric key predistribution techniques.
-
2026-09-23 Cubical Sheaf Complexes with Constant Expansion with Applications to Asymptotically Good qLTCs.
Summary: For every fixed integers $r \ge 4$ and $2 \le k \le r-2$, we construct $r$-dimensional cubical sheaf complexes whose degree-$k$ CSS codes have positive constant rate, linear distance, and constant soundness, with bounded row and column weights. Taking $r=4$ and $k=2$ gives a family of asymptotically good binary qLTCs. At the core of our construction is a uniform product-expansion theorem for explicit Reed-Solomon codes on norm-one evaluation sets. The key point is that the expansion constant stays bounded away from zero as the local code lengths grow. We place these codes on arithmetic cubical complexes, obtaining constant local expansion for both the resulting sheaf and its dual. Together with the local-to-global framework of Dinur, Lin, and Vidick (FOCS 2024) and sheaf duality, this gives linear distance and constant soundness, while an asymmetric choice of local code dimensions gives positive rate. The resulting codes are explicit and polynomial-time computable.
-
2026-09-23 New Applications of RSA.
Summary: We present new constructions of the following primitives featuring succinct communication: - Batch threshold encryption: Assuming RSA and DCR, we construct a threshold encryption scheme in which a batch of ciphertexts can be decrypted via short hints from a quorum of decryptors. Previously, this was known only from bilinear maps. - Aggregate signatures: Assuming RSA, we construct a signature scheme where signatures on different messages from an arbitrary number of signers can be aggregated into one signature. Our construction achieves bounded-query security. In contrast to prior (fully secure) schemes in the standard model, ours makes black-box use of cryptography. - Broadcast encryption: Assuming RSA, we construct a secret-key encryption scheme where a sender can encrypt a message to multiple recipients via a short ciphertext. Our construction achieves bounded-query security. Previously, (fully secure) broadcast encryption was known from bilinear maps or obfustopia or nonstandard lattice assumptions. Our constructions are simple and achieve optimal succinctness in the key parameter (decryption batch size or number of signers or recipient set size). In this metric, they outperform or match known constructions. Our results are obtained by applying the lens of projective pseudorandom generators (PPRGs) to these problems. Towards this, we identify new enhanced properties for PPRGs and instantiate them from the RSA assumption.
-
2026-09-22 Good Quantum Locally Testable Codes from Product Expansion.
Summary: We construct quantum locally testable codes (LTCs) with constant rate, distance, soundness and locality under a variant of a product expansion conjecture of Bafna and Vyas about Reed-Solomon codes. In particular, we use the high-dimensional expansion framework of Dinur, Lin and Vidick for constructing quantum LTCs, instantiated with the non-Abelian cubical complexes of Rungtanapirom, Stix and Vdovina. Our code is obtained by equipping the complex with carefully chosen Reed-Solomon local codes whose symmetries are compatible with those of the complex.
-
2026-09-22 Strong Selective and List-Decoding Direct Product Theorems for Quantum Query Complexity.
Summary: Quantum strong direct-product theorems for specific functions have been known for nearly two decades. These have been extended to general results for function computation and state generation. The proofs of these results use a version of the multiplicative adversary method that does not naturally extend to relations. Standard strong direct-product theorems apply when algorithms must correctly answer every given question. Prior work extended them to equivalent threshold direct-product theorems, which require answers to all questions but only require that most answers are correct. We focus on two further generalizations. Strong selective direct-products apply to algorithms that adaptively choose, based on what they learn from queries, which questions from a large list to answer. This generalization is relational and useful for proving time-space tradeoffs. We prove a quantum strong selective direct-product theorem for all functions using a relational formulation of the multiplicative adversary method by Jeffery and Zur which we prove, via an equivalent formulation, satisfies a strong selective direct product property and captures any query lower bound for functions proven by negative-weights adversaries. The second generalization is list-decoding direct product problems introduced by Ben-David and Blais for classical query complexity. These allow an algorithm to produce a large list of possible output vectors such that one of them is fully correct. They proved that such theorems hold for randomized complexity of all Boolean functions. We prove a quantum analogue of this theorem for all partial Boolean-valued functions. We show that strong list-decoding direct-product theorems are implied by a special case of multiplicative adversaries which we show, via a new reduction, can be obtained from negative-weights adversaries for any Boolean-valued function.
-
2026-09-20 Cryptanalysis of LDVR: Polynomial-Time Adaptive Attacks against Threshold Schnorr Signatures.
Summary: We present the first polynomial-time attacks against certain parameter choices of the low-dimensional vector representation (LDVR) problem, introduced by Crites, Katz, Komlo, Tessaro, and Zhu (CRYPTO '25) to prove the adaptive security of FROST in the Algebraic Group Model. Building on work by Crites and Stewart (CRYPTO '25), they also show that every attack against LDVR directly yields an adaptive attack—i.e., one leveraging adaptive corruptions—against any threshold Schnorr signature that exposes to the attacker the values $g^{\mathrm{sk}_i}$ for the secret-key shares $\mathrm{sk}_i$ held by the signers, where $g$ is a generator of the underlying group. This class of protocols includes FROST (Komlo and Goldberg, SAC '20), Sparkle (Crites, Komlo, and Maller, CRYPTO '23), Lindell's three-round protocol (CiC '24), along with other masking-free protocols. Consequently, our results give the first polynomial-time adaptive attacks against a class of practical threshold signatures. Our attacks require a large, but still polynomial, number of signers: asymptotically, we require $n = \Theta(\log^2 p)$ signers and corrupt $t_c = \Theta(\log p)$ of them. For the prime-order subgroup of Curve25519, concrete and optimized instantiations of our attack require $n=5020$ signers and corrupt $204$ of them. This exceeds the signer counts commonly encountered in practice, but falls within the scenarios considered by NIST's call for threshold signatures. At the theoretical level, our results also yield a non-artificial separation between static and adaptive security based on a real-world multi-party protocol. Our results arise from a new reduction from LDVR to structured subset-sum problems determined by the evaluation points of the polynomial used to secret-share the signing key. We further show that allowing these evaluation points to be chosen adversarially can lead to substantially more efficient attacks.
-
2026-09-18 Batched and Weighted Threshold Encryption with Silent Setup.
Summary: Batched Threshold Encryption (BTE) enables a committee of parties to jointly decrypt any subset of ciphertexts from a large set, while all other ciphertexts remain private. BTE has applications in blockchains, particularly in designing encrypted mempools, where transactions are encrypted until included into a block to prevent maximal extractable value (MEV) attacks. Existing BTE constructions, however, encounter one of the following usability downsides: (a) reliance on expensive distributed key generation (DKG) protocols, (b) dependency on a batch label during ciphertext generation, (c) ciphertexts that are only decryptable if they are encoded with respect to a non-conflicting index in the batch, or (d) no support for weighted threshold policies. In this work, we present a BTE scheme that simultaneously addresses all of the above issues. It has a silent setup (key generation does not require interaction between parties), is batch and index independent, and supports weighted threshold policies. Our construction is based on prime-order groups with bilinear pairings with security proof in the Generic Group Model. Our ciphertext size is also the shortest among existing silent batched threshold encryption schemes with only $2$ group elements in addition to the message length.
-
2026-09-17 An $m^{2.943}$ Bohnenblust--Hille Bound on the Boolean Cube.
Summary: Let $q_m=2m/(m+1)$ and put \[ β_0=\frac{3}{2}+\frac{1}{\log 2}=2.9426950408\ldots, \] where $\log$ is the natural logarithm. We give a proof scheme showing that, for every $\varepsilon>0$, there is $C_\varepsilon<\infty$ such that every complex-valued function $f:\{-1,1\}^n\to\C$ of Fourier degree at most $m$ satisfies \[ \left(\sum_{A\subseteq[n]}\abs{\wh f(A)}^{q_m}\right)^{1/q_m} \le C_\varepsilon m^{β_0+\varepsilon}\norm{f}_\infty. \] The improvement over the $m^9$ estimate of the earlier draft has two ingredients. The first three Fourier levels are estimated at the scales $m$, $m^{3/2}$, and $m^{5/3}$. These losses are encoded in the weight $M^{μ_r/r}r^B\min\{r,L_M\}^8$, $L_M\asymp_B \log (M+1)$, with $μ_r=\lceil3r/2\rceil$ for $r\ge2$. A parity-compatible central window handles $r
-
2026-09-15 Quantum Encodings, Private Messages, and Communication Complexity.
Summary: We study the complexity of quantum decomposable randomized encodings (QDRE). We establish a bi-directional connection between the QDRE model and the communication complexity model of quantum private simultaneous message protocols (QPSM), where classical communication is free, and quantum communication is the primary complexity measure. We show the following upper and lower bounds. (1) For every quantum channel mapping $n$ to $m$ qubits, we give a QPSM protocol with quantum communication $m$, using exponential pre-shared entanglement. In particular, constant-output channels have $O(1)$ quantum communication, and classical-output channels require no quantum communication. (2) With $O(n)$ pre-shared entanglement, every one-qubit-output channel has an $O(n)$-quantum-communication QPSM protocol. Conversely, there exists a channel and a constant $c>0$ for which any protocol with at most $cn$ pre-shared entanglement requires $Ω(n)$ quantum communication. (3) In contrast, we identify a structured class of quantum channels, Clifford-induced channels, for which there exist QPSM protocols with $O(n)$ pre-shared entangled qubits and only $O(1)$ quantum communication complexity. Through our connection, we obtain analogous upper and lower bounds on the quantum encoding size of QDRE. Our results show that the amount of pre-shared entanglement plays a central role in quantum encoding size and quantum communication complexity.
-
2026-09-11 Rank-1-perturbed trickledown theorems: Mixing time of Glauber dynamics for the Sherrington-Kirkpatrick model up to $β\leq \frac{1}{2}+\varepsilon$.
Summary: We introduce a new family of trickledown theorems, a.k.a., local to global technique to bound the spectral gap of the Glauber dynamics for multi-state spin systems. In this technique instead of upper-bounding the influence matrix of a link of co-dimension 2 by $λI$ (where $λ$ is the second eigenvalue of the link), we upper-bound the influence matrix after a carefully chosen rank-1 shift. The rank-1 shift allows for a significantly smaller upper-bound but it comes at the cost of bounding the average loss due to rank-1 perturbations. As an application we use this method to show that the natural Glauber dynamics mixes in polynomial time to generate samples from the Sherrington-Kirkpatrick model for $β\leq \tfrac{1}{2}+\varepsilon$, for an absolute constant $\varepsilon>0$. At the heart of the proof we manage to bound the loss due to rank-1 perturbations by averaging over all links of co-dimension 2.
-
2026-09-07 Polynomial Bohnenblust--Hille bounds for product of cyclic groups.
Summary: Fix an integer $K\ge2$, and let $C_K^n =\{(e^{\frac{2πij}{K}})_{j=0}^{K-1}\}^n$ be the product of cyclic groups of order $K$. For a Fourier character $χ_α$, let $s(α)$ be the number of active coordinates. We give a self-contained proposed proof that the dimension-free Bohnenblust--Hille constants governed by interaction order grow polynomially: if $p_d=2d/(d+1)$ and \[ γ_2=\frac12, \qquad γ_K=\frac{K\log(K-1)}{4(K-2)}\quad(K\ge3), \] then the $\ell^{p_d}$ norm of Fourier coefficients $\{\hat f(α)\}$ is bounded by $L^\infty$ norm of $f$ multiplied by $C(K) d^{4γ_K+5}$. The constant $C(K)$ is actually at most of the order $K^{5/2}$.
-
2026-09-05 The $I_{3322}$ Bell inequality requires infinite dimensions.
Summary: The $I_{3322}$ inequality is one of the simplest bipartite Bell inequalities, with three possible questions and two possible answers per party. Yet, despite its simplicity, it has long been conjectured by Pál and Vértesi (Physical Review A, 2010) to possess the following quirk: no finite-dimensional quantum strategy can attain its maximal violation, but an infinite-dimensional strategy can. A Bell correlation, with five questions and three answers per party, that provably possesses the same property has since been discovered by Coladangelo and Stark (Nature Communications, 2020). However, proving that the same property holds for $I_{3322}$, which lives in the simplest Bell scenario in which such a phenomenon could occur, has remained elusive. Here, we provide a proof of this conjecture. An approximate proof, containing all the main ideas, was produced by GPT 5.5 Pro after various rounds of interaction. The presentation of this proof was revised substantially by the author for completeness, correctness, and clarity.
-
2026-09-02 A quantum oracle separation between QMA(2) and QMA.
Summary: We find a quantum oracle relative to which $\mathsf{QMA} \neq \mathsf{QMA}(2)$. As a consequence, we resolve the no-disentanglers conjecture of Watrous: for every $ε+δ<1$, any $(ε,δ)$-disentangler requires input size exponential in the number of output qubits. Our proof combines the unitarily invariant polynomial method of She and Yuen (ITCS '23) with a new construction based on the symmetric and antisymmetric subspace projectors, reducing the $\mathsf{QMA}$ lower bound to the approximate degree of $\mathrm{OR}$.
-
2026-08-31 Unconditional Certified Randomness without Structure.
Summary: We obtain a certified randomness protocol in the quantum random oracle model. The protocol is non-interactive and publicly verifiable with a classical verifier, and is based on Yamakawa and Zhandry's proof of quantumness [JACM'24]. We prove unconditional security of this protocol against adversaries making subexponentially-many adaptive quantum queries to the random oracle. Prior work on certified randomness relative to a random oracle additionally assumed the Aaronson--Ambainis conjecture or proved security only against low query-depth adversaries.
-
2026-08-30 A note on the $Σ_2^P$-completeness of the Frobenius number.
Summary: Given a finite set $A$ of natural numbers whose greatest common divisor is one, the Frobenius number $g(A)$ is the largest integer that is not a non-negative integer combination of the numbers in $A$. In a 2016 preprint, Matsubara states that given $A$ and $k$, deciding if $g(A) \geq k$ is $Σ_2^P$-complete. A decade has passed since without peer-reviewed publication of this result. At the same time, the community has found it difficult to verify this result. In this note, we give a write-up of the completeness proof based on Matsubara (2016).
-
2026-08-18 Unclonable encryption from BB84 states: a simultaneous Goldreich-Levin reduction.
Summary: Goldreich-Levin reductions are ubiquitous in cryptography: they convert an algorithm capable of guessing $\langle r, m \rangle$ (mod $2$) for a hidden string $m$ and a random challenge $r$, to one that is capable of extracting the entirety of $m$. Here, we describe a "simultaneous" Goldreich-Levin reduction for two entangled parties who are capable of guessing $\langle r, m \rangle$ given uniformly random identical challenges $r$. This allows to upgrade any unclonable encryption scheme satisfying "search" security to one satisfying the gold standard of unclonable "indistinguishability". As a corollary, we show that the simplest candidate unclonable encryption scheme from BB84 states satisfies unclonable indistinguishability. This result was discovered by GPT-5.6 Ultra after a few interactions. Our prompts included recent results on unclonable encryption by Ananth and Sahai, and Ragavan.
-
2026-08-11 Optimal von Neumann Entropy Estimation.
Summary: We show that the von Neumann entropy of an unknown $d$-dimensional quantum state $ρ$ can be estimated to within additive error $\varepsilon$ using \[ Θ\!\left(\frac{d^2}{\varepsilon\log^2(d)(1+\varepsilon\log^2(d))} + \frac{\log^2(d)}{\varepsilon^2}\right) \] samples of $ρ$, matching the lower bound due to Wang (2026). Our estimator is based on a high-degree approximation polynomial and the main technical challenge is bounding its variance despite the large polynomial degree. We overcome this using a quantum Hoeffding decomposition, which preserves cancellations between different polynomial terms and reduces the analysis to derivative bounds for the approximation polynomial.
-
2026-08-07 Tight Security for BBS Signatures.
Summary: This paper studies the concrete security of BBS signatures (Boneh, Boyen, Shacham, CRYPTO '04; Camenisch and Lysyanskaya, CRYPTO '04), a popular algebraic construction of digital signatures which underlies practical privacy-preserving authentication systems and is undergoing standardization by the W3C and IRTF. Schäge (Journal of Cryptology '15) gave a tight standard-model security proof under the q-SDH assumption for a less efficient variant of the scheme, called BBS+--here, q is the number of issued signatures. In contrast, the security proof for BBS (Tessaro and Zhu, EUROCRYPT '23), also under the q-SDH assumption, is \emph{not} tight. Nonetheless, this recent proof shifted both standardization and industry adoption towards the more efficient BBS, instead of BBS+, and for this reason, it is important to understand whether this tightness gap is inherent. Recent cryptanalysis by Chairattana-Apirom and Tessaro (ASIACRYPT '25) also shows that a tight reduction to q-SDH is the best we can hope for. This paper closes this gap in two different ways. On the positive end, we show a novel tight reduction for BBS in the case where each message is signed at most once--this case covers in particular the common practical use case which derandomizes signing. On the negative end, we use a meta-reduction argument to prove that if we allow generating multiple signatures for the same message, then {\em no} algebraic reduction to q-SDH (and its variants) can be tight.
-
2026-08-05 Tightness of and counterexamples to several quantum estimates.
Summary: We prove here several tightness results for such quantum inequalities as 1) the comparison of operator norm and product norm of $d$-local hamiltonians, 2) Bohnenblust--Hille inequality for $d$-local hamiltonians, 3) for quantum Fourier entropy-influence conjecture. We give a deterministic prove of the sharpness of Bohnenblust--Hille exponent. We also discuss the quantum Aaronson--Ambainis conjecture in a special case of anti-commuting Pauli strings.
-
2026-08-04 Attribute-based Undetectable Watermarking for Generative AI Models.
Summary: Generative AI systems increasingly produce content whose provenance is difficult to verify, motivating watermarking techniques for identifying model-generated outputs. Existing cryptographic watermarking methods provide strong undetectability guarantees: without a detection key, watermarked outputs are computationally indistinguishable from unwatermarked ones. However, these approaches do not address the crucial deployment challenge of how to safely delegate detection capabilities. With an unrestricted detection key, a malicious detector may use the detection key beyond its intended scope, enabling watermark sanitization, scope abuse, and user profiling. To mitigate this safety concern, we introduce, to the best of our knowledge, the first \emph{attribute-based watermarking} for generative AI models, providing fine-grained, policy-controlled watermark detection. In our approach, each generated output is associated with attributes, and each detection key is \emph{constrained by a policy} on potential attributes. A detection key can only be used to detect watermarked outputs whose attributes satisfy the corresponding policy, while watermarked outputs that fall outside the policy remain computationally indistinguishable from unwatermarked ones. We construct such an attribute-based watermarking scheme and formalize its security properties, including consistency, adaptive robustness to bounded corruptions, undetectability, and soundness, along with a security proof under standard cryptographic assumptions. Our construction integrates constrained pseudorandom functions, pseudorandom error-correcting codes, and randomness recovery procedures with generative AI models. Finally, we implement a prototype and an empirical evaluation, demonstrating that attribute-based watermarking is both effective and practical.
-
2026-08-03 Beyond the PPAD hardness of Auto-bidding Auctions.
Summary: Computing certain autobidding equilibria is PPAD complete in the worst case. Yet such instances rarely arise in practice, where advertisers running simple, decentralized learning strategies usually converge quickly. We show there is no contradiction: the hardness requires atomicity and vanishes once the value distribution is nonatomic, as it is in real world markets. To bridge worst case hardness and practical convergence, we introduce diffuse analysis, a beyond worst case framework that studies equilibrium computation when bidder values are drawn from general nonatomic distributions. Under this framework, the autobidding equilibrium becomes a separately monotone generalized Nash equilibrium (GNE). For this GNE, we give the first solver with last iterate linear convergence. Thus, the equilibrium has polynomial diffuse complexity, matching the convergence observed in real-world markets. Concretely, our framework subsumes the budget pacing and the throttling equilibrium as special cases when the payment rule is a convex combination of first and second price.
-
2026-08-02 Dense Hamiltonians at the Parseval Limit: The Noncommutative BH Constant is Exponential and the Quantum FEI Conjecture is False.
Summary: For each $d\geq 1$ we construct a norm-1 Hermitian operator whose Pauli expansion contains $N(d)=\exp(Ω(d^2))$ terms, each of degree $d$ and magnitude $1/\sqrt{N(d)}$ - the largest magnitude permitted by Parseval's identity. For comparison, if a bounded diagonal operator (or equivalently, a bounded degree-$d$ function on the Boolean cube) has $N(d)$ Pauli coefficients, all of magnitude $Ω(1/\sqrt{N(d)})$, then $N(d)\leq \exp(\widetilde{O}(d^{1.5}))$. This construction implies the noncommutative Bohnenblust--Hille (BH) constant satisfies $\mathrm{BH}_{M_2}(d)\geq\exp(Ω(d))$. Together with the upper bounds proved in prior work, this settles the asymptotic growth of $\mathrm{BH}_{M_2}(d)$ as exponential. Our lower bound also asymptotically separates $\mathrm{BH}_{M_2}(d)$ from the (classical) hypercube BH constant $\mathrm{BH}_{\{\pm 1\}}(d)$, which in turn is known to be subexponential: $\mathrm{BH}_{\{\pm 1\}}(d)\leq C^{\sqrt{d \log d}}$. Our Hamiltonians are also unitary and thus quantum Boolean functions in the sense of Montanaro and Osborne (2010). As such they refute the quantum Fourier Entropy-Influence conjecture of Bu et al. (2024), a natural generalization of the classical Fourier Entropy-Influence conjecture due to Friedgut and Kalai (1996).
-
2026-07-29 Robust quantum state certification and uncertainty principles for total influence.
Summary: We show that nonadaptive single-qubit Pauli measurements suffice to test whether an unknown $n$-qubit state $ρ$ is $\varepsilon$-close to or $O(\varepsilon)$-far from an ideal target state $|ψ\rangle$, for all but a $2^{-Ω(n)}$ fraction of target states. The test uses $O(\varepsilon^{-2}\log(1/δ))$ copies of $ρ$ to achieve confidence $1-δ$, which is information-theoretically optimal even among protocols with arbitrary joint measurements. The main technical innovation is an uncertainty principle for weighted generalizations of the total influence of Boolean functions. As a simple example, the unweighted variant states that $\mathbf{Inf}[f]+\mathbf{Inf}[\widehat{f}] = Ω(n)$, which is a natural hypercube analogue of the Heisenberg uncertainty principle (here $\widehat{\,\cdot\,}$ denotes the $2^{-n/2}$-normalized Fourier transform). The weighted case generalizes $\mathbf{Inf}[\,\cdot\,]$ and $\mathbf{Inf}[\,\widehat{\,\cdot\,}\,]$ to Dirichlet energies associated with Glauber dynamics for certain dual measures on the cube.
-
2026-07-25 Defense via Consensus: Mitigating Subliminal Learning through Multi-Teacher Distillation.
Summary: Fine-tuning can introduce unwanted behaviors through deliberate poisoning or unexpected generalization from seemingly benign data. Because these effects may be difficult to anticipate or detect, we seek robustness in the fine-tuning process itself. Our premise is that datasets independently collected for the same task would share the intended capability while differing in their unwanted effects. This creates an opportunity to learn from agreement across sources while reducing dependence on any individual source. We introduce Multi-Source Consensus Distillation (MSCD): we fine-tune one model on each dataset, use agreement among these models to generate replacement training responses, and train a fresh student on the resulting data. We develop minimum and base-relative rules for combining the models' predictions, with extensions that accommodate useful behaviors learned by only some models or expressed in different words. Experiments on controlled poisoning, subliminal learning, and emergent misalignment show suppression of unwanted behavior while retaining shared benefits. In a practical language-understanding task, MSCD retains intent-classification capability while reducing unsafe medical advice introduced during fine-tuning. Comparing distilled students with direct consensus decoding shows that distillation closely preserves the measured desirable and undesirable behavior rates while replacing multiple model evaluations per generated token with a single student evaluation.
-
2026-06-26 Linear-size $\ell_1$ sparsifiers.
Summary: We prove that for any matrix $A \in \mathbb{R}^{m \times n}$ and any $\varepsilon > 0$ there is a diagonal matrix $D \in \mathbb{R}_{\geq 0}^{m \times m}$ with at most $O(\frac{n}{\varepsilon^2})$ nonzero entries so that \[ (1-\varepsilon) \|Ax\|_1 \leq \|DAx\|_1 \leq (1+\varepsilon)\|Ax\|_1 \quad \forall x \in \mathbb{R}^n. \] In particular, for any zonotope $Z \subseteq \mathbb{R}^{n}$ there exists a zonotope $Z' \subseteq \mathbb{R}^{n}$ generated by at most $O(\frac{n}{\varepsilon^2})$ segments so that $(1-\varepsilon) Z \subseteq Z' \subseteq (1+\varepsilon) Z$. Previously, the best known bound was $O(\frac{n}{\varepsilon^2} \log n)$ due to Talagrand (1990).
-
2026-06-23 No linear stretching of single-copy quantum pseudorandomness.
Summary: Pseudorandom states, introduced by Ji, Liu, and Song (CRYPTO 2018), are quantum analogues of classical pseudorandom generators. A fundamental property of classical pseudorandom generators is that their output can be stretched to arbitrary polynomial length. Whether an analogous stretching property holds for quantum pseudorandom states has been elusive. Here, we essentially resolve this question for the single-copy variant of pseudorandom states, 1PRS. We show, via a unitary oracle separation, that no fully black-box construction can stretch the output length of a 1PRS by any constant factor greater than one.
-
2026-06-09 Density estimation for Hellinger via minimum-distance estimators: mixtures of Gaussians, log-concave, and more.
Summary: We study the task of density estimation, where we hope to accurately estimate a probability density from $n$ samples. A textbook method for density estimation in total variation distance is the minimum-distance estimator approach, where we conclude both the algorithm and the analysis merely from bounding the VC dimension of a particular concept class (the so-called Yatracos class). While this technique has originally yielded sharp guarantees primarily for total variation distance, in this work we extend the minimum-distance estimator approach for learning within Hellinger distance. Our main observation is that we may produce an analogous recipe for Hellinger (where we only require bounding the VC dimension of a related concept class) by drawing connections to recent results yielding reverse data processing inequalities. This recipe is flexible enough to accommodate fast algorithms originally designed for total variation distance; by modifying the approach of Acharya et al. (2017) we conclude the first near-linear time algorithm for learning classes including univariate mixtures of log-concave densities and mixtures of Gaussians (with arbitrary variances), with near-optimal sample complexity.
-
2026-05-31 On Thin Perfect Matchings up to Polylogarithmic Factors.
Summary: We resolve the thin matching problem proposed by Anari, Charikar and Ramakrishnan [ACR23] up to polylogarithmic factors. Given a fractional perfect matching $x$, we say a perfect matching $M$ is $α$-thin w.r.t. $x$ if for any cut $(S,\overline{S})$, we have $$ |M \cap E(S,\overline{S})| \leq α\cdot x(S,\overline{S}).$$ [ACR23] conjectured that for any fractional perfect matching $x$, there exists a perfect matching $M$ which is $O(1)$-thin w.r.t. $x$. First, we show that if $M$ is restricted to be in the support of $x$, then $α\geq Ω(n)$ and we complement this by designing an efficient algorithm that outputs an $O(n\log n)$-thin perfect matching where $n$ is the number of vertices. Then, we relax this constraint and show that for any fractional perfect matching $x$, there is a perfect matching $M$ (which is not necessarily in the support of $x$) such that $M$ is $\text{polylog}(n)$-thin w.r.t. $x$. All results work for both bipartite and non-bipartite graphs. We also discuss applications to the metric distortion problem.
-
2026-05-29 High-Dimensional Expanders, the Sparsest Cut Problem, and Steurer's Conjecture.
Summary: In 2010, Steurer conjectured that any family of $n$ unit-norm vectors $v_1,\dots,v_n$ with polynomially small average correlation $\mathbb{E}_{i,j}|\langle v_i,v_j\rangle|\leq n^{-ε}$ contains linear-sized constant-separated sets. We refute this conjecture in a strong sense using the machinery of sparse high-dimensional expanders: such vector families do not even have linear-sized $\frac{1}{\log^{1/4-o(1)}(n)}$-separated sets. Consequently, we show that there are families of vertex expanders on $n$ vertices for which the (average) $L_2$-mixing time to the uniform distribution of any reweighted simple random walk is at least $\log^{5/4-o(1)} n$.
-
2026-05-14 Extending CDCL to disjunctions of parity equations.
Summary: Because CDCL produces proofs in the Resolution proof system, problems provably hard for Resolution are also provably hard for CDCL. Exponentially shorter proofs can sometimes be found using stronger proof systems such as $\text{Res}(\oplus)$, a generalization of Resolution to XNF formulas, whose constraints are disjunctions of parity equations ("linear clauses") such as $(x \oplus y) \lor \lnot (y \oplus z)$. While some modern solvers like CryptoMiniSAT reason on Boolean clauses with separate parity equations, reasoning about more general linear clauses is less explored. We present $\text{CDCL}(\oplus)$, a generalization of CDCL to XNF formulas, and prove a bidirectional connection with $\text{Res}(\oplus)$: $\text{CDCL}(\oplus)$ not only produces $\text{Res}(\oplus)$ proofs, but also polynomially simulates $\text{Res}(\oplus)$ given nondeterministic decisions and restarts, mirroring the classical relationship between CDCL and Resolution. Our key technical tool is a new set of inference rules for $\text{Res}(\oplus)$ that helps us translate Resolution-based subroutines such as 1-UIP clause learning. Altogether, $\text{CDCL}(\oplus)$'s parity reasoning includes branching on arbitrary parity equations, linear-algebraic reasoning during unit propagation, and learning linear clauses through conflict analysis. We provide a proof-of-concept implementation of $\text{CDCL}(\oplus)$ called Xorcle, which includes adaptations of existing CDCL heuristics to XNF formulas and an extension of LRUP proof logging that we call $\text{LRUP}(\oplus)$. On a selected suite of benchmarks focusing on native XNF formulas, Xorcle outperforms existing solvers such as Kissat and CryptoMiniSAT. Additionally, on Tseitin formulas written in CNF, even without preprocessing, Xorcle's running time appears to scale nearly polynomially.
-
2026-05-10 Instance-Adaptive Online Multicalibration.
Summary: We study online multicalibration beyond the worst-case. We give a single, efficient algorithm which dynamically interpolates between benign and worst-case sequences by adaptively refining a dyadic grid of prediction values. Its error is controlled by the number of leaves in the refinement tree. Our analysis recovers the known $\widetilde O(T^{2/3})$ worst-case-optimal rate for online multicalibration, while simultaneously automatically adapting to easier instances: in the marginal stochastic setting it obtains a rate of $\widetilde O(\sqrt T)$, and for piecewise-stationary means with $J$ segments its rate is $\widetilde O(\sqrt{JT})$. More generally, the rate depends on a threshold-complexity measure of the predictable mean process relative to the group family. We show that this dependence is tight up to logarithmic factors.
-
2026-05-07 Weak-to-Strong Generalization is Nearly Inevitable (in Linear Models).
Summary: Weak-to-strong generalization is a phenomenon in post-training whereby a strong student model, when finetuned solely with feedback from a weaker teacher, can not only surpass the teacher, but can improve upon its own capabilities. Recent work of Burns et al. (2023) demonstrated that this can occur in the setting of frontier language models, and subsequently there has been a flurry of both empirical work trying to exploit this phenomenon, as well as theoretical work attempting to understand it. In this work, we demonstrate that weak-to-strong generalization occurs in standard linear logistic regression, under mild distributional assumptions on the data. In fact, we show that this happens for most student-teacher pairs, suggesting that weak-to-strong generalization is in fact \emph{almost inevitable}, even in this basic setting. Notably, our setting does not require the student to be more expressive or have more model capacity in any way compared to the teacher, which runs contrary to the prevailing theoretical belief that a mismatch in model capacity is a central mechanism to weak-to-strong generalization.
-
2026-05-05 Nearly-Tight Bounds for Zonotope Containment and Beyond.
Summary: We investigate the convex-body containment problem $\max\{s >0 : s Z \subseteq Q\}$, where the outer body $Q \subseteq \mathbb R^d$ is described by a membership oracle and the inner body $Z \subseteq \mathbb R^d$ is a zonotope. Our main result is a sampling-based $O(\sqrt{d})$-approximation algorithm for this problem that almost matches the lower bound of $Ω(\sqrt{d/\log d})$ by Khot and Naor in the oracle model. Assuming zonotopes can be sparsified by a linear number of generators, which is referred to as Talagrand conjecture, our approach attains the optimal approximation factor of $Θ(\sqrt{d/\log d})$. Our second main result is a proof of Talagrand's conjecture for $Δ$-modular zonotopes whenever $Δ$ is constant. Those zonotopes are of the form $Z = \{ Wx \colon \| x\|_\infty \leq 1\}$ where the non-zero $d \times d$ sub-determinants of $W$ are between $1$ and $Δ$. This result establishes a connection between zonoid sparsification and spectral sparsification of Batson, Spielman and Srivastava. We complement these results with a universal $Ω(\sqrt{d/\log d})$ lower bound holding for all zonotopes. Finally, we consider containment problems $\max\{s >0 : s K \subseteq Q\}$, for general convex bodies $K \subseteq \mathbb R^d$. A result of Naszódi on approximating $K \subseteq \mathbb R^d$ by a polytope implies a $Θ(d/\log d)$ approximation algorithm in polynomial time. We show the tightness of this approximation factor in the oracle model via a reduction to the circumradius computation. Our lower bound holds for centrally symmetric convex sets, implying that Barvinok's optimal $O(\sqrt{d})$-approximation of a centrally symmetric convex body by a polytope with a polynomial number of vertices cannot be computed in polynomial time.
-
2026-04-23 Sampling from the Hardcore Model on Random Regular Bipartite Graphs above the Uniqueness Threshold.
Summary: We design an efficient sampling algorithm to generate samples from the hardcore model on random regular bipartite graphs as long as $λ\lesssim \frac{1}{\sqrtΔ}$, where $Δ$ is the degree. Combined with recent work of Jenssen, Keevash and Perkins this implies an FPRAS for the partition function of the hardcore model on random regular bipartite graphs at any fugacity. Our algorithm is shown by analyzing two new Markov chains that work in complementary regimes. Our proof then proceeds by showing the corresponding simplicial complexes are top-link spectral expanders and appealing to the trickle-down theorem to prove fast mixing.
-
2026-04-09 The Boolean surface area of polynomial threshold functions.
Summary: Polynomial threshold functions (PTFs) are an important low-complexity class of Boolean functions, with strong connections to learning theory and approximation theory. Recent work on learning and testing PTFs has exploited structural and isoperimetric properties of the class, especially bounds on average sensitivity, one of the central themes in the study of PTFs since the Gotsman--Linial conjecture. In this work we study PTFs through the lens of the Boolean surface area (or Talagrand boundary) \[ \mathbf{BSA}[f]=\mathbb{E}|\nabla f|=\mathbb{E}\sqrt{s_{f}(x)}, \] a natural measure of vertex-boundary complexity on the discrete cube. Our main result is that every degree-$d$ PTF has polylogarithmic Boolean surface area: \[ \mathbf{BSA}[f]\le C_d(\log(en))^{C_d}. \] The proof is based on the PTF Restriction Lemma of Kabanets, Kane, and Lu \cite{KKL2017} and proceeds through a tail bound for the pointwise sensitivity. In particular, it controls all subcritical fractional moments of the sensitivity. We also record a random block partition principle for Boolean surface area and an alternative recursive argument following Kane's work \cite{DK} on average sensitivity, which independently yields the weaker bound \[ \mathbf{BSA}[f]\le \exp(C_d\sqrt{\log n}). \]
-
2026-02-18 Separating Oblivious and Adaptive Models of Variable Selection.
Summary: Sparse recovery is among the most well-studied problems in learning theory and high-dimensional statistics. In this work, we investigate the statistical and computational landscapes of sparse recovery with $\ell_\infty$ error guarantees. This variant of the problem is motivated by \emph{variable selection} tasks, where the goal is to estimate the support of a $k$-sparse signal in $\mathbb{R}^d$. Our main contribution is a provable separation between the \emph{oblivious} (``for each'') and \emph{adaptive} (``for all'') models of $\ell_\infty$ sparse recovery. We show that under an oblivious model, the optimal $\ell_\infty$ error is attainable in near-linear time with $\approx k\log d$ samples, whereas in an adaptive model, $\gtrsim k^2$ samples are necessary for any algorithm to achieve this bound. This establishes a surprising contrast with the standard $\ell_2$ setting, where $\approx k \log d$ samples suffice even for adaptive sparse recovery. We conclude with a preliminary examination of a \emph{partially-adaptive} model, where we show nontrivial variable selection guarantees are possible with $\approx k\log d$ measurements.
-
2026-02-17 Unforgeable Watermarks for Language Models via Robust Signatures.
Summary: Language models now routinely produce text that is difficult to distinguish from human writing, raising the need for robust tools to verify content provenance. Watermarking has emerged as a promising countermeasure, with existing work largely focused on model quality preservation and robust detection. However, current schemes provide limited protection against false attribution. We strengthen the notion of soundness by introducing two novel guarantees: unforgeability and recoverability. Unforgeability prevents adversaries from crafting false positives, texts that are far from any output from the watermarked model but are nonetheless flagged as watermarked. Recoverability provides an additional layer of protection: whenever a watermark is detected, the detector identifies the source text from which the flagged content was derived. Together, these properties strengthen content ownership by linking content exclusively to its generating model, enabling secure attribution and fine-grained traceability. We construct the first undetectable watermarking scheme that is robust, unforgeable, and recoverable with respect to substitutions (i.e., perturbations in Hamming metric). The key technical ingredient is a new cryptographic primitive called robust (or recoverable) digital signatures, which allow verification of messages that are close to signed ones, while preventing forgery of messages that are far from all previously signed messages. We show that any standard digital signature scheme can be boosted to a robust one using property-preserving hash functions (Boyle, LaVigne, and Vaikuntanathan, ITCS 2019).
-
2026-02-12 The Power of Two Bases: Robust and copy-optimal certification of nearly all quantum states with few-qubit measurements.
Summary: A central task in quantum information science is state certification: testing whether an unknown state is $ε_1$-close to a fixed target state, or $ε_2$-far. Recent work has shown that surprisingly simple measurement protocols--comprising only single-qubit measurements--suffice to certify arbitrary $n$-qubit states [Huang, Preskill, Soleimanifar '25; Gupta, He, O'Donnell '25]. However, these certification protocols are not robust: rather than allowing constant $ε_1$, they can only positively certify states within $ε_1=O(1/n)$ trace distance of the target. In many experimental settings, the appropriate error tolerance is constant as the system size grows, so this lack of robustness renders existing tests inapplicable at scale, no matter how many times the test is repeated. Here we present robust certification protocols based on few-qubit measurements that apply to all but a $O(2^{-n})$-fraction of pure target states. Our first protocol achieves constant robustness, i.e. $ε_1=Θ(1)$, using a single $O(\log n)$-qubit measurement along with single-qubit measurements in the $Z$ or $X$ basis on the other qubits. As a corollary of its robustness, this protocol also achieves constant (in $n$) copy complexity, which is optimal. Our second protocol uses exclusively single-qubit measurements and is nearly robust: $ε_1=Ω(1/\log n)$. Our tests are based on a new uncertainty principle for conditional fidelities, which may be of independent interest.
-
2026-01-13 Obfuscation of Arbitrary Quantum Circuits (via Subspace-Preserving Pseudorandom Unitary).
Summary: Program obfuscation aims to conceal a computer program's internal structure while preserving its functionality. A central open problem is whether an obfuscation scheme for arbitrary quantum circuits exists. Despite several efforts toward this goal, prior works have succeeded only in obfuscating quantum circuits that implement either pseudo-deterministic functions or unitary transformations. Although unitary transformations already include a broad class of quantum computations, many important quantum tasks, such as state preparation, quantum sampling, and quantum error correction, go beyond unitaries and are described by general completely positive trace-preserving (CPTP) maps. In this work, we construct the first quantum ideal obfuscation scheme for arbitrary quantum circuits that support quantum inputs and outputs in the classical oracle model assuming post-quantum one-way functions, thereby resolving an open problem raised in Bartusek et al. (STOC 2023), Bartusek, Brakerski, and Vaikuntanathan (STOC 2024), and Huang and Tang (FOCS 2025). At the core of our construction lies a novel primitive that we introduce, called the subspace-preserving strong pseudorandom unitary (spsPRU). An spsPRU is a family of efficient unitaries that fix every vector in a given linear subspace $S$, while acting as a Haar random unitary on its orthogonal complement $S^\perp$ under both forward and inverse oracle queries. Furthermore, by instantiating the classical oracle model with the ideal obfuscation scheme for classical circuits proposed by Jain et al. (CRYPTO 2023) and later enhanced by Bartusek et al. (CRYPTO 2026), our obfuscation scheme can also be realized in the quantumly accessible pseudorandom oracle model.
-
2026-01-09 Rigorous Implications of the Low-Degree Heuristic.
Summary: Over the past decade, the low-degree heuristic has been used to estimate the algorithmic thresholds for a wide range of average-case planted vs null distinguishing problems. Such results rely on the hypothesis that if the low-degree moments of the planted and null distributions are sufficiently close, then no efficient (noise-tolerant) algorithm can distinguish between them. This hypothesis is appealing due to the simplicity of calculating the low-degree likelihood ratio (LDLR) -- a quantity that measures the similarity between low-degree moments. However, despite sustained interest in the area, it remains unclear whether low-degree indistinguishability actually rules out any interesting class of algorithms. In this work, we initiate the study and develop technical tools for translating LDLR upper bounds to rigorous lower bounds against concrete algorithms. As a consequence, we prove: for any permutation-invariant distribution $\mathsf{P}$, 1. If $\mathsf{P}$ is over $\{0,1\}^n$ and is low-degree indistinguishable from $U = \mathrm{Unif}(\{0,1\}^n)$, then a noisy version of $\mathsf{P}$ is statistically indistinguishable from $U$. 2. If $\mathsf{P}$ is over $\mathbb{R}^n$ and is low-degree indistinguishable from the standard Gaussian ${N}(0, 1)^n$, then no statistic based on symmetric polynomials of degree at most $O(\log n/\log \log n)$ can distinguish between a noisy version of $\mathsf{P}$ from ${N}(0, 1)^n$. 3. If $\mathsf{P}$ is over $\mathbb{R}^{n\times n}$ and is low-degree indistinguishable from ${N}(0,1)^{n\times n}$, then no constant-sized subgraph statistic can distinguish between a noisy version of $\mathsf{P}$ and ${N}(0, 1)^{n\times n}$.
-
2026-01-07 Exact Bounds for Forbidden Configurations and the Extremal Matrices.
Summary: Let $F$ be a $k\times \ell$ (0,1)-matrix. A matrix is simple if it is a (0,1)-matrix with no repeated columns. A (0,1)-matrix $A$ is said to have a $F$ as a configuration if there is a submatrix of $A$ which is a row and column permutation of $F$. In the language of sets, a configuration is a trace. Let $\mathrm{Avoid}(m,F)$ be all simple $m$-rowed matrices $A$ with no configuration $F$. Define $\mathrm{forb}(m,F)$ as the maximum number of columns of any matrix in $\mathrm{Avoid}(m,F)$. The $2\times (p+1)$ (0,1)-matrix $F(0,p,1,0)$ consists of a row of $p$ 1's and a row of one 1 in the remaining column. The paper determines $\mathrm{forb}(m,F(0,p,1,0))$ for $1\le p\le 9$ and the extremal matrices are characterized. A construction may be extremal for all $p$.
-
2025-12-07 Ideal Attribution and Faithful Watermarks for Language Models.
Summary: We introduce ideal attribution mechanisms, a formal abstraction for reasoning about attribution decisions over strings. At the core of this abstraction lies the ledger, an append-only log of the prompt-response interaction history between a model and its user. Each mechanism produces deterministic decisions based on the ledger and an explicit selection criterion, making it well-suited to serve as a ground truth for attribution. We frame the design goal of watermarking schemes as faithful representation of ideal attribution mechanisms. This novel perspective brings conceptual clarity, replacing piecemeal probabilistic statements with a unified language for stating the guarantees of each scheme. It also enables precise reasoning about desiderata for future watermarking schemes, even when no current construction achieves them, since the ideal functionalities are specified first. In this way, the framework provides a roadmap that clarifies which guarantees are attainable in an idealized setting and worth pursuing in practice.
-
2025-11-21 High-Accuracy List-Decodable Mean Estimation.
Summary: In list-decodable learning, we are given a set of data points such that an $α$-fraction of these points come from a nice distribution $D$, for some small $α\ll 1$, and the goal is to output a short list of candidate solutions, such that at least one element of this list recovers some non-trivial information about $D$. By now, there is a large body of work on this topic; however, while many algorithms can achieve optimal list size in terms of $α$, all known algorithms must incur error which decays, in some cases quite poorly, with $1 / α$. In this paper, we ask if this is inherent: is it possible to trade off list size with accuracy in list-decodable learning? More formally, given $ε> 0$, can we can output a slightly larger list in terms of $α$ and $ε$, but so that one element of this list has error at most $ε$ with the ground truth? We call this problem high-accuracy list-decodable learning. Our main result is that non-trivial high-accuracy guarantees, both information-theoretically and algorithmically, are possible for the canonical setting of list-decodable mean estimation of identity-covariance Gaussians. Specifically, we demonstrate that there exists a list of candidate means of size at most $L = \exp \left( O\left( \tfrac{\log^2 1 / α}{ε^2} \right)\right)$ so that one of the elements of this list has $\ell_2$ distance at most $ε$ to the true mean. We also design an algorithm that outputs such a list with runtime and sample complexity $n = d^{O(\log L)} + \exp \exp (\widetilde{O}(\log L))$. We do so by demonstrating a completely novel proof of identifiability, as well as a new algorithmic way of leveraging this proof without the sum-of-squares hierarchy, which may be of independent technical interest.
-
2025-11-12 Separating QMA from QCMA with a classical oracle.
Summary: We construct a classical oracle proving that, in a relativized setting, the set of languages decidable by an efficient quantum verifier with a quantum witness (QMA) is strictly bigger than those decidable with access only to a classical witness (QCMA). The separating classical oracle we construct is for a decision problem we coin spectral Forrelation -- the oracle describes two subsets of the boolean hypercube, and the computational task is to decide if there exists a quantum state whose standard basis measurement distribution is well supported on one subset while its Fourier basis measurement distribution is well supported on the other subset. This is equivalent to estimating the spectral norm of a "Forrelation" matrix between two sets that are accessible through membership queries. Our lower bound derives from a simple observation that a query algorithm with a classical witness can be run multiple times to generate many samples from a distribution, while a quantum witness is a "use once" object. This observation allows us to reduce proving a QCMA lower bound to proving a sampling hardness result which does not simultaneously prove a QMA lower bound. To prove said sampling hardness result for QCMA, we observe that quantum access to the oracle can be compressed by expressing the problem in terms of bosons -- a novel "second quantization" perspective on compressed oracle techniques, which may be of independent interest. Using this compressed perspective on the sampling problem, we prove the sampling hardness result, completing the proof.
-
2025-11-10 How Do Data Owners Say No? A Case Study of Data Consent Mechanisms in Web-Scraped Vision-Language AI Training Datasets.
Summary: The internet has become the main source of data to train modern text-to-image or vision-language models, yet it is increasingly unclear whether web-scale data collection practices for training AI systems adequately respect data owners' wishes. Ignoring the owner's indication of consent around data usage not only raises ethical concerns but also has recently been elevated into lawsuits around copyright infringement cases. In this work, we aim to reveal information about data owners' consent to AI scraping and training, and study how it's expressed in DataComp, a popular dataset of 12.8 billion text-image pairs. We examine both the sample-level information, including the copyright notice, watermarking, and metadata, and the web-domain-level information, such as a site's Terms of Service (ToS) and Robots Exclusion Protocol. We estimate at least 122M of samples exhibit some indication of copyright notice in CommonPool, and find that 60\% of the samples in the top 50 domains come from websites with ToS that prohibit scraping. Furthermore, we estimate 9-13\% with 95\% confidence interval of samples from CommonPool to contain watermarks, where existing watermark detection methods fail to capture them in high fidelity. Our holistic methods and findings show that data owners rely on various channels to convey data consent, of which current AI data collection pipelines do not entirely respect. These findings highlight the limitations of the current dataset curation/release practice and the need for a unified data consent framework taking AI purposes into consideration.
-
2025-11-06 Optimal Inference Schedules for Masked Diffusion Models.
Summary: A major bottleneck of standard auto-regressive large language models is that their inference process is inherently sequential, resulting in very long and costly inference times. To circumvent this, practitioners proposed a class of language models called diffusion language models, of which the masked diffusion model (MDM) is the most successful. The MDM is able to sample tokens out-of-order and, ostensibly, many tokens at once and in parallel. However, there is very limited rigorous understanding of how much parallel sampling these models can perform without noticeable degradation in their sampling performance. Prior work of Li and Cai obtained some preliminary bounds, but these are not tight for many natural classes of distributions. In this work, we give a new, exact characterization of the expected divergence between the true distribution and the sampled distribution, for any distribution and any unmasking schedule for the sampler, showing an elegant connection to the theory of univariate function approximation. By leveraging this connection, we then attain a number of novel lower and upper bounds for this problem. While the connection to function approximation in principle gives the optimal unmasking schedule for any distribution, we show that it is in general impossible to compete with it without strong a priori knowledge of the distribution, even in seemingly benign settings. However, we also demonstrate new upper bounds and new sampling schedules in terms of well-studied information-theoretic properties of the base distribution, namely, its total correlation and dual total correlation, which show that in some natural settings, one can sample in $O(log n)$ steps without any visible loss in performance, where $n$ is the total sequence length.
-
2025-11-04 Spectral Certificates and Non-commutative Sum-of-Squares Lower Bounds for Hamiltonians.
Summary: A central question in quantum many-body physics is estimating the ground energy of a $k$-local Hamiltonian system. In this work, we present a spectral technique for certifying a lower bound on the ground energy of a random $n$-qubit Hamiltonian system defined as the sum of signed $k$-local Pauli operators. In particular, we prove that for any constant $\ell$, there exists an efficiently computable length $n^{O(\ell)}$ certificate that is always a lower bound on the ground energy with the promise that, with high probability over the random Hamiltonian distribution, the certificate value is an $\varepsilon$-good approximation of the true ground energy when the number of terms is sufficiently large. Second, we show by construction that this technique, while successful on average over random Hamiltonian systems, can fail to produce good certificates on worst-case instances. Our spectral technique for producing these certificates comes from extending classical results on $k$-XOR refutations to $k$-local Pauli Hamiltonians by crafting a quantum variant of the Kikuchi matrix for CSP refutations. To show the limitations of this technique, we prove non-commutative Sum-of-Squares lower bounds for worst-case signed $k$-local Pauli operators. More generally, we explore how the non-commutative Sum-of-Squares relaxation can be understood as augmenting the standard Sum-of-Squares relaxation with the commutation relations between the Pauli operators. We instantiate the resulting framework with a modification to prior quantum code-based NLTS Hamiltonians that yields stronger complexity guarantees for the low-energy space; our Hamiltonian family satisfies simultaneously (1) $Ω(\log n)$-circuit depth lower bounds for all low-energy states, (2) constant-gap NP-hardness to approximate the ground energy, and (3) a constant-gap non-commutative Sum-of-Squares integrality gap up to $Ω(n)$-levels.
-
2025-10-23 Excluding a Line Minor via Design Matrices and Column Number Bounds for the Circuit Imbalance Measure.
Summary: For a real matrix $A \in \mathbb{R}^{d \times n}$ with non-collinear columns, we show that $n \leq O(d^4 κ_A)$ where $κ_A$ is the \emph{circuit imbalance measure} of $A$. The circuit imbalance measure $κ$ is a real analogue of $Δ$-modularity for integer matrices, satisfying $κ_A \leq Δ_A$ for integer $A$. The circuit imbalance measure has numerous applications in the context of linear programming (see Ekbatani, Natura and V{é}gh (2022) for a survey). Our result generalizes the $O(d^4 Δ_A)$ bound of Averkov and Schymura (2023) for integer matrices and provides the first polynomial bound holding for all parameter ranges on real matrices. To derive our result, similar to the strategy of Geelen, Nelson and Walsh (2021) for $Δ$-modular matrices, we show that real representable matroids induced by $κ$-bounded matrices are minor closed and exclude a rank $2$ uniform matroid on $O(κ)$ elements as a minor (also known as a line of length $O(κ)$). As our main technical contribution, we show that any simple rank $d$ complex representable matroid which excludes a line of length $l$ has at most $O(d^4 l)$ elements. This complements the tight bound of $(l-3)\binom{d}{2} + d$ for $l \geq 4$, of Geelen, Nelson and Walsh which holds when the rank $d$ is sufficiently large compared to $l$ (at least doubly exponential in $l$).
-
2025-10-09 Agnostic Product Mixed State Tomography via Robust Statistics.
Summary: We study the complexity of two closely related learning problems, one quantum and one classical. In the quantum setting, we consider agnostic tomography for the natural class of product mixed states. Given $N$ copies of an $n$-qubit state $ρ$, the goal is to output a nearly optimal product mixed state approximation in trace distance. While recent work has focused on pure-state ansatz (e.g., product or stabilizer states), no polynomial-time guarantees were previously known for mixed-state ansatz. In the classical setting, we study robust learning of binary product distributions: given samples from an unknown distribution on ${0,1}^n$, the goal is to output a nearly optimal product approximation. Our main contributions are as follows. (1) We give a semi-agnostic tomography algorithm for product mixed states with polynomial sample and computational complexity achieving error $O(\mathrm{opt}\log(1/\mathrm{opt}))$, where $\mathrm{opt}$ is the trace distance to the best product approximation. This is the first efficient algorithm with any nontrivial agnostic guarantee for mixed-state ansatz, using only single-qubit, single-copy measurements. We also prove a Quantum Statistical Query lower bound showing near-optimality, and an unconditional lower bound demonstrating that adaptivity is necessary under single-qubit measurements. (2) We give a semi-agnostic algorithm for robustly learning binary product distributions with matching guarantees and establish a Statistical Query lower bound, essentially resolving the efficient robust learnability of this class and improving on prior work since Diakonikolas et al. (2016).
-
2025-10-09 A Meta-Complexity Characterization of Minimal Quantum Cryptography.
Summary: We give a meta-complexity characterization of EFI pairs, which are considered the "minimal" primitive in quantum cryptography (and are equivalent to quantum commitments). More precisely, we show that the existence of EFI pairs is equivalent to the following: there exists a non-uniformly samplable distribution over pure states such that the problem of estimating a certain Kolmogorov-like complexity measure is hard given a single copy. A key technical step in our proof, which may be of independent interest, is to show that the existence of EFI pairs is equivalent to the existence of non-uniform single-copy secure pseudorandom state generators (nu 1-PRS). As a corollary, we get an alternative, arguably simpler, construction of a universal EFI pair.
-
2025-10-08 Trickle-down Theorems via C-Lorentzian Polynomials II: Pairwise Spectral Influence and Improved Dobrushin's Condition.
Summary: Let $μ$ be a probability distribution on a multi-state spin system on a set $V$ of sites; equivalently, a $d$-partite simplicial complex with distribution $μ$ on maximal faces. For any pair of vertices $u,v\in V$, define the pairwise spectral influence $\mathcal{I}_{u,v}$ as follows. Let $σ$ be a choice of spins $s_w\in S_w$ for every $w\in V\setminus\{u,v\}$, and construct a matrix in $\mathbb{R}^{(S_u\cup S_v)\times (S_u\cup S_v)}$ where for any $s_u\in S_u, s_v\in S_v$, the $(us_u,vs_v)$-entry is the probability that $s_v$ is the spin of $v$ conditioned on $s_u$ being the spin of $u$ and on $σ$. Then $\mathcal{I}_{u,v}$ is the maximal second eigenvalue of this matrix, over all choices of spins for all $w\in V\setminus\{u,v\}$. Equivalently, $\mathcal{I}_{u,v}$ is the maximum local spectral expansion of links of codimension $2$ that include a spin for every $w \in V \setminus \{u,v\}$. We show that if the largest eigenvalue of the pairwise spectral influence matrix with entries $\mathcal{I}_{u,v}$ is bounded away from 1, i.e. $λ_{\max}(\mathcal{I})\leq 1-ε$ (and $X$ is connected), then the Glauber dynamics mixes rapidly and generate samples from $μ$. This improves/generalizes the classical Dobrushin's influence matrix as the $\mathcal{I}_{u,v}$ lower-bounds the classical influence of $u\to v$. As an application, we prove that the Glauber dynamics mixes rapidly up to (approximately) the phase transition for the multi-state hardcore model--a widely studied model in telecommunication networks and statistical physics (generalizing the hardcore model) introduced by Mazel and Suhov. As a by-product of our results, we also prove improved/almost optimal trickle-down theorems for partite simplicial complexes. Our proof builds on the trickle-down theorems via $\mathcal{C}$-Lorentzian polynomials machinery recently developed by the authors and Lindberg.
-
2025-10-06 Quantum precomputation: parallelizing cascade circuits and the Moore-Nilsson conjecture is false.
Summary: Parallelization is a major challenge in quantum algorithms due to physical constraints like no-cloning. This is vividly illustrated by the conjecture of Moore and Nilsson from their seminal work on quantum circuit complexity [MN01, announced 1998]: unitaries of a deceptively simple form--controlled-unitary "staircases"--require circuits of minimum depth $Ω(n)$. If true, this lower bound would represent a major break from classical parallelism and prove a quantum-native analogue of the famous NC $\neq$ P conjecture. In this work we settle the Moore-Nilsson conjecture in the negative by compressing all circuits in the class to depth $O(\log n)$, which is the best possible. The parallelizations are exact, ancilla-free, and can be computed in poly($n$) time. We also consider circuits restricted to 2D connectivity, for which we derive compressions of optimal depth $O(\sqrt{n})$. More generally, we make progress on the project of quantum parallelization by introducing a quantum blockwise precomputation technique somewhat analogous to the method of Arlazarov, Dinič, Kronrod, and Faradžev [Arl+70] in classical dynamic programming, often called the "Four-Russians method." We apply this technique to more-general "cascade" circuits as well, obtaining for example polynomial depth reductions for staircases of controlled $\log(n)$-qubit unitaries.
-
2025-09-26 T-TAMER: Provably Taming Trade-offs in ML Serving.
Summary: As machine learning models continue to grow in size and complexity, efficient serving faces increasingly broad trade-offs spanning accuracy, latency, resource usage, and other objectives. Multi-model serving further complicates these trade-offs; for example, in cascaded models, each early-exit decision balances latency reduction against potential accuracy loss. Despite the pervasiveness and importance of such trade-offs, current strategies remain largely heuristic and case-specific, limiting both their theoretical guarantees and general applicability. We present a general framework, T-Tamer, which formalizes this setting as a multi-stage decision process, where the objective is to determine both when to exit and which model to consult. Our main result shows that recall (i.e., the ability to revisit earlier models) is both necessary and sufficient for achieving provable performance guarantees. In particular, we prove that strategies without recall cannot obtain any constant-factor approximation to the optimal trade-off, whereas recall-based strategies provably attain the optimal trade-off in polynomial time. We validate our analysis through experiments on synthetic datasets and early-exit workloads for vision and NLP benchmarks. The results show that recall-based strategies consistently yield efficient accuracy-latency trade-offs. We hope this work provides a principled foundation for bridging heuristic practice with theoretical guarantees in the design of early-exit and cascaded models.
-
2025-09-25 Semi-Random Graphs, Robust Asymmetry, and Reconstruction.
Summary: The Graph Reconstruction Conjecture famously posits that any undirected graph on at least three vertices is determined up to isomorphism by its family of (unlabeled) induced subgraphs. At present, the conjecture admits partial resolutions of two types: 1) casework-based demonstrations of reconstructibility for families of graphs satisfying certain structural properties, and 2) probabilistic arguments establishing reconstructibility of random graphs by leveraging average-case phenomena. While results in the first category capture the worst-case nature of the conjecture, they play a limited role in understanding the general case. Results in the second category address much larger graph families, but it remains unclear how heavily the necessary arguments rely on optimistic distributional properties. Drawing on the perspectives of smoothed and semi-random analysis, we study the robustness of what are arguably the two most fundamental properties in this latter line of work: asymmetry and uniqueness of subgraphs. Notably, we find that various semi-random graph distributions exhibit these properties asymptotically, much like their Erdős-Rényi counterparts. In particular, Bollobás (1990) demonstrated that almost all Erdős-Rényi random graphs $G = (V, E) \sim \mathscr{G}(n, p)$ enjoy the property that their induced subgraphs on $n - Θ(1)$ vertices are asymmetric and mutually non-isomorphic, for $1 - p, p = Ω(\log(n) / n)$. We show that this property is robust against perturbation -- even when an adversary is permitted to add/remove each vertex pair in $V^{(2)}$ with (independent) arbitrarily large constant probability. Exploiting this result, we derive asymptotic characterizations of asymmetry in random graphs with planted structure and bounded adversarial corruptions, along with improved bounds on the probability mass of nonreconstructible graphs in $\mathscr{G}(n, p)$.
-
2025-09-25 Lower Bounds for Learning Hamiltonians from Time Evolution.
Summary: Learning about a Hamiltonian $H$ from its time evolution $e^{-iHt}$ is a fundamental task in quantum science. A flurry of recent work has developed powerful new algorithms with provable guarantees for this task, for a variety of natural settings. Despite this, relatively little is known about lower bounds for learning Hamiltonians. In particular, in the natural setting where we assume $H$ is a $k$-local Hamiltonian on $n$ qubits, all existing algorithms require total evolution time at least $n^{Ω(k)}$ to learn the parameters of $H$, and it remained open whether one could obtain even faster algorithms -- or at the very least, if one could obtain better runtimes for simpler tasks, such as estimating a single designated coefficient of the Hamiltonian. In this work we show the answer is essentially no, by obtaining strong lower bounds for these problems. We find that not only do $k$-local Hamiltonians require $n^{Ω(k)}$ time evolution or interactions to learn, but also that in several senses, learning anything about a Hamiltonian is just as hard as learning everything. In particular, we find the same $n^{Ω(k)}$ lower bound holds for learning a single coefficient of a $k$-local Hamiltonian $H$, even if the rest of $H$ is already known. We also show an $n^{Ω(k)}$ lower bound for the task of effective Hamiltonian learning, where one seeks only to learn a unitary that approximately implements time evolution of $H$. Several related lower bounds, such as for general sparse (but not necessarily local) $H$ are also given. On the technical side, we make a new connection between Hamiltonian learning lower bounds and the analysis of Boolean functions, where we introduce a novel extremal property that may be of independent interest.
-
2025-09-21 Auditability and the Landscape of Distance to Multicalibration.
Summary: Calibration is a critical property for establishing the trustworthiness of predictors that provide uncertainty estimates. Multicalibration is a strengthening of calibration which requires that predictors be calibrated on a potentially overlapping collection of subsets of the domain. As multicalibration grows in popularity with practitioners, an essential question is: how do we measure how multicalibrated a predictor is? Błasiok et al. (2023) considered this question for standard calibration by introducing the distance to calibration framework (dCE) to understand how calibration metrics relate to each other and the ground truth. Building on the dCE framework, we consider the auditability of the distance to multicalibration of a predictor $f$. We begin by considering two natural generalizations of dCE to multiple subgroups: worst group dCE (wdMC), and distance to multicalibration (dMC). We argue that there are two essential properties of any multicalibration error metric: 1) the metric should capture how much $f$ would need to be modified in order to be perfectly multicalibrated; and 2) the metric should be auditable in an information theoretic sense. We show that wdMC and dMC each fail to satisfy one of these two properties, and that similar barriers arise when considering the auditability of general distance to multigroup fairness notions. We then propose two (equivalent) multicalibration metrics which do satisfy these requirements: 1) a continuized variant of dMC; and 2) a distance to intersection multicalibration, which leans on intersectional fairness desiderata. Along the way, we shed light on the loss-landscape of distance to multicalibration and the geometry of the set of perfectly multicalibrated predictors. Our findings may have implications for the development of stronger multicalibration algorithms as well as multigroup auditing more generally.
-
2025-09-18 The Story of Sunflowers.
Summary: Sunflowers, or $Δ$-systems, are a fundamental concept in combinatorics introduced by Erdős and Rado in their paper: {\em Intersection theorems for systems of sets}, J. Lond. Math. Soc. (1) {\bf 35} (1960), 85--90. A sunflower is a collection of sets where all pairs have the same intersection. This paper explores the wide-ranging applications of sunflowers in computer science and combinatorics. We discuss recent progress towards the sunflower conjecture and present a short elementary proof of the best known bounds for the robust sunflower lemma.
-
2025-09-15 Approximating the operator norm of local Hamiltonians via few quantum states.
Summary: Consider a Hermitian operator $A$ acting on a complex Hilbert space of dimension $2^n$. We show that when $A$ has small degree in the Pauli expansion, or in other words, $A$ is a local $n$-qubit Hamiltonian, its operator norm can be approximated independently of $n$ by maximizing $|\braket{ψ|A|ψ}|$ over a small collection $\mathbf{X}_n$ of product states $\ketψ\in (\mathbf{C}^{2})^{\otimes n}$. More precisely, we show that whenever $A$ is $d$-local, \textit{i.e.,} $°(A)\le d$, we have the following discretization-type inequality: \[ \|A\|\le C(d)\max_{ψ\in \mathbf{X}_n}|\braket{ψ|A|ψ}|. \] The constant $C(d)$ depends only on $d$. This collection $\mathbf{X}_n$ of $ψ$'s, termed a \emph{quantum norm design}, is independent of $A$, and consists of product states, and can have cardinality as small as $(1+\eps)^n$, which is essentially tight. Previously, norm designs were known only for homogeneous $d$-localHamiltonians $A$ \cite{L,BGKT,ACKK}, and for non-homogeneous $2$-local traceless $A$ \cite{BGKT}. Several other results, such as boundedness of Rademacher projections for all levels and estimates of operator norms of random Hamiltonians, are also given.
-
2025-09-01 The curious case of "XOR repetition" of monogamy-of-entanglement games.
Summary: In this work, we consider "decision" variants of a monogamy-of-entanglement game by Tomamichel, Fehr, Kaniewski, and Wehner [New Journal of Physics '13]. In its original "search" variant, Alice prepares a (possibly entangled) state on registers $\mathsf{ABC}$; register $\mathsf{A}$, consisting of $n$ qubits, is sent to a Referee, while $\mathsf{B}$ and $\mathsf{C}$ are sent to Bob and Charlie; the Referee then measures each qubit in the standard or Hadamard basis (chosen uniformly at random). The basis choices are sent to Bob and Charlie, whose goal is to simultaneously guess the Referee's $n$-bit outcome string $x$. Tomamichel et al. show that the optimal winning probability is $\cos^{2n} {(\fracπ{8})}$, following a perfect parallel repetition theorem. We consider the following "decision" variants of this game: - Variant 1, "XOR repetition": Bob and Charlie's goal is to guess the XOR of all the bits of $x$. Ananth et al. [Asiacrypt '24] conjectured that the optimal advantage over random guessing decays exponentially in $n$. Surprisingly, we show that this conjecture is false, and, in fact, there is no decay at all: there exists a strategy that wins with probability $\cos^2{(\fracπ{8})} \approx 0.85$ for any $n$. - Variant 2, "Goldreich-Levin": The Referee additionally samples a uniformly random $n$-bit string $r$ that is sent to Bob and Charlie along with the basis choices. Their goal is to guess the parity of $r\cdot x$. We show that the optimal advantage over random guessing decays exponentially in $n$ for the restricted class of adversaries that do not share entanglement. A similar result was already shown by Champion et al. and Çakan et al.; we give a more direct proof. Additionally, we put forward a reasonably concrete conjecture that is equivalent to exponential decay for general adversaries.
-
2025-08-25 Spectral Refutations of Semirandom $k$-LIN over Larger Fields.
Summary: We study the problem of strongly refuting semirandom $k$-LIN$(\mathbb{F})$ instances: systems of $k$-sparse inhomogeneous linear equations over a finite field $\mathbb{F}$. For the case of $\mathbb{F} = \mathbb{F}_2$, this is the well-studied problem of refuting semirandom instances of $k$-XOR, where the works of [GKM22,HKM23] establish a tight trade-off between runtime and clause density for refutation: for any choice of a parameter $\ell$, they give an $n^{O(\ell)}$-time algorithm to certify that there is no assignment that can satisfy more than $\frac{1}{2} + \varepsilon$-fraction of constraints in a semirandom $k$-XOR instance, provided that the instance has $O(n) \cdot \left(\frac{n}{\ell}\right)^{k/2 - 1} \log n /\varepsilon^4$ constraints, and the work of [KMOW17] provides good evidence that this tight up to a $\mathrm{polylog}(n)$ factor via lower bounds for the Sum-of-Squares hierarchy. However for larger fields, the only known results for this problem are established via black-box reductions to the case of $\mathbb{F}_2$, resulting in an $|{\mathbb{F}}|^{3k}$ gap between the current best upper and lower bounds. In this paper, we give an algorithm for refuting semirandom $k$-LIN$(\mathbb{F})$ instances with the "correct" dependence on the field size $|{\mathbb{F}}|$. For any choice of a parameter $\ell$, our algorithm runs in $(|{\mathbb{F}}|n)^{O(\ell)}$-time and strongly refutes semirandom $k$-LIN$(\mathbb{F})$ instances with at least $O(n) \cdot \left(\frac{|{\mathbb{F}^*}| n}{\ell}\right)^{k/2 - 1} \log(n |{\mathbb{F}^*}|) /\varepsilon^4$ constraints. We give good evidence that this dependence on the field size $|{\mathbb{F}}|$ is optimal by proving a lower bound for the Sum-of-Squares hierarchy that matches this threshold up to a $\mathrm{polylog}(n |{\mathbb{F}^*}|)$ factor. Our results also extend to the more general case of finite Abelian groups.
-
2025-08-20 Robust Estimation Under Heterogeneous Corruption Rates.
Summary: We study the problem of robust estimation under heterogeneous corruption rates, where each sample may be independently corrupted with a known but non-identical probability. This setting arises naturally in distributed and federated learning, crowdsourcing, and sensor networks, yet existing robust estimators typically assume uniform or worst-case corruption, ignoring structural heterogeneity. For mean estimation for multivariate bounded distributions and univariate gaussian distributions, we give tight minimax rates for all heterogeneous corruption patterns. For multivariate gaussian mean estimation and linear regression, we establish the minimax rate for squared error up to a factor of $\sqrt{d}$, where $d$ is the dimension. Roughly, our findings suggest that samples beyond a certain corruption threshold may be discarded by the optimal estimators -- this threshold is determined by the empirical distribution of the corruption rates given.
-
2025-08-14 Welfare-Centric Clustering.
Summary: Fair clustering has traditionally focused on ensuring equitable group representation or equalizing group-specific clustering costs. However, Dickerson et al. (2025) recently showed that these fairness notions may yield undesirable or unintuitive clustering outcomes and advocated for a welfare-centric clustering approach that models the utilities of the groups. In this work, we model group utilities based on both distances and proportional representation and formalize two optimization objectives based on welfare-centric clustering: the Rawlsian (Egalitarian) objective and the Utilitarian objective. We introduce novel algorithms for both objectives and prove theoretical guarantees for them. Empirical evaluations on multiple real-world datasets demonstrate that our methods significantly outperform existing fair clustering baselines.
-
2025-07-25 Forbidden Configurations and Boundary Cases.
Summary: Let $F$ be a $k\times \ell$ (0,1)-matrix. Define a (0,1)-matrix $A$ to have a $F$ as a \emph{configuration} if there is a submatrix of $A$ which is a row and column permutation of $F$. In the language of sets, a configuration is a \emph{trace}. Define a matrix to be {\it simple} if it is a (0,1)-matrix with no repeated columns. Let $\mathrm{Avoid}(m,F)$ be all simple $m$-rowed matrices $A$ with no configuration $F$. Define $\mathrm{forb}(m,F)$ as the maximum number of columns of any matrix in $\mathrm{Avoid}(m,F)$. Determining $\mathrm{forb}(m,F)$ requires determining bounds and constructions of matrices in $\mathrm{Avoid}(m,F)$. The paper considers some column maximal $k$-rowed simple $F$ that have the bound $Θ(m^{k-2})$ and yet adding a column increases bound to $Ω(m^{k-1})$. By a construction, $\mathrm{forb(m,F)}$ is determined exactly.
-
2025-07-16 Obfuscation of Unitary Quantum Programs.
Summary: Program obfuscation aims to hide the inner workings of a program while preserving its functionality. In the quantum setting, recent works have obtained obfuscation schemes for specialized classes of quantum circuits. For instance, Bartusek, Brakerski, and Vaikuntanathan (STOC 2024) constructed a quantum state obfuscation scheme, which supports the obfuscation of quantum programs represented as quantum states for pseudo-deterministic quantum programs with classical inputs and outputs in the classical oracle model. In this work, we improve upon existing results by constructing the first quantum state obfuscation scheme for unitary (or approximately unitary) quantum programs supporting quantum inputs and outputs in the classical oracle model. At the core of our obfuscation scheme are two novel ingredients: a functional quantum authentication scheme that allows key holders to learn specific functions of the authenticated quantum state with simulation-based security, and a compiler that represents an arbitrary quantum circuit as a projective linear-plus-measurement quantum program described by a sequence of non-adaptive Clifford gates interleaved with adaptive and compatible measurements.
-
2025-07-08 The Delta Learning Hypothesis: Preference Tuning on Weak Data can Yield Strong Gains.
Summary: Improvements in language models are often driven by improving the quality of the data we train them on, which can be limiting when strong supervision is scarce. In this work, we show that paired preference data consisting of individually weak data points can enable gains beyond the strength of each individual data point. We formulate the delta learning hypothesis to explain this phenomenon, positing that the relative quality delta between points suffices to drive learning via preference tuning--even when supervised finetuning on the weak data hurts. We validate our hypothesis in controlled experiments and at scale, where we post-train 8B models on preference data generated by pairing a small 3B model's responses with outputs from an even smaller 1.5B model to create a meaningful delta. Strikingly, on a standard 11-benchmark evaluation suite (MATH, MMLU, etc.), our simple recipe matches the performance of Tulu 3, a state-of-the-art open model tuned from the same base model while relying on much stronger supervisors (e.g., GPT-4o). Thus, delta learning enables simpler and cheaper open recipes for state-of-the-art post-training. To better understand delta learning, we prove in logistic regression that the performance gap between two weak teacher models provides useful signal for improving a stronger student. Overall, our work shows that models can learn surprisingly well from paired data that might typically be considered weak.
-
2025-06-30 Sampling and Identity-Testing Without Approximate Tensorization of Entropy.
Summary: Certain tasks in high-dimensional statistics become easier when the underlying distribution satisfies a local-to-global property called approximate tensorization of entropy (ATE). For example, the Glauber dynamics Markov chain of an ATE distribution mixes fast and can produce approximate samples in a small amount of time, since such a distribution satisfies a modified log-Sobolev inequality. Moreover, identity-testing for an ATE distribution requires few samples if the tester is given coordinate conditional access to the unknown distribution, as shown by Blanca, Chen, Štefankovič, and Vigoda (COLT 2023). A natural class of distributions that do not satisfy ATE consists of mixtures of (few) distributions that do satisfy ATE. We study the complexity of identity-testing and sampling for these distributions. Our main results are the following: 1. We show fast mixing of Glauber dynamics from a data-based initialization, with optimal sample complexity, for mixtures of distributions satisfying modified log-Sobolev inequalities. This extends work of Huang, Koehler, Lee, Mohanty, Rajaraman, Vuong, and Wu (STOC 2025, COLT 2025) for mixtures of distributions satisfying Poincaré inequalities. 2. Answering an open question posed by Blanca et al., we give efficient identity-testers for mixtures of ATE distributions in the coordinate-conditional sampling access model. We also give some simplifications and improvements to the original algorithm of Blanca et al.
-
2025-06-28 MPC in the Quantum Head (or: Superposition-Secure (Quantum) Zero-Knowledge).
Summary: The MPC-in-the-head technique (Ishai et al., STOC 2007) is a celebrated method to build zero-knowledge protocols with desirable theoretical properties and high practical efficiency. This technique has generated a large body of research and has influenced the design of real-world post-quantum cryptographic signatures. In this work, we present a generalization of the MPC-in-the-head paradigm to the quantum setting, where the MPC is running a quantum computation. As an application of our framework, we propose a new approach to build zero-knowledge protocols where security holds even against a verifier that can obtain a superposition of transcripts. This notion was pioneered by Damgard et al., who built a zero-knowledge protocol for NP (in the common reference string model) secure against superposition attacks, by relying on perfectly hiding and unconditionally binding dual-mode commitments. Unfortunately, no such commitments are known from standard cryptographic assumptions. In this work we revisit this problem, and present two new three-round protocols in the common reference string model: (i) A zero-knowledge argument for NP, whose security reduces to the standard learning with errors (LWE) problem. (ii) A zero-knowledge argument for QMA from the same assumption.
-
2025-06-20 A Common Pool of Privacy Problems: Legal and Technical Lessons from a Large-Scale Web-Scraped Machine Learning Dataset.
Summary: We investigate the contents of web-scraped data for training AI systems, at sizes where human dataset curators and compilers no longer manually annotate every sample. Building off of prior privacy concerns in machine learning models, we ask: What are the legal privacy implications of web-scraped machine learning datasets? In an empirical study of a popular training dataset, we find significant presence of personally identifiable information despite sanitization efforts. Our audit provides concrete evidence to support the concern that any large-scale web-scraped dataset may contain legally defined personal data. We use these findings of a real-world dataset to inform our legal analysis with respect to existing privacy and data protection laws. We surface various legal risks of current data curation practices that may propagate personal information to train downstream models. Based on our empirical and legal analyses, we argue for reorientation of current frameworks of "publicly available" information to meaningfully limit the development of AI built upon indiscriminate scraping of the internet.
-
2025-06-05 Improved Regret Bounds for Linear Bandits with Heavy-Tailed Rewards.
Summary: We study stochastic linear bandits with heavy-tailed rewards, where the rewards have a finite $(1+ε)$-absolute central moment bounded by $\upsilon$ for some $ε\in (0,1]$. We improve both upper and lower bounds on the minimax regret compared to prior work. When $\upsilon = \mathcal{O}(1)$, the best prior known regret upper bound is $\tilde{\mathcal{O}}(d T^{\frac{1}{1+ε}})$. While a lower with the same scaling has been given, it relies on a construction using $\upsilon = \mathcal{O}(d)$, and adapting the construction to the bounded-moment regime with $\upsilon = \mathcal{O}(1)$ yields only a $Ω(d^{\fracε{1+ε}} T^{\frac{1}{1+ε}})$ lower bound. This matches the known rate for multi-armed bandits and is generally loose for linear bandits, in particular being $\sqrt{d}$ below the optimal rate in the finite-variance case ($ε= 1$). We propose a new elimination-based algorithm guided by experimental design, which achieves regret $\tilde{\mathcal{O}}(d^{\frac{1+3ε}{2(1+ε)}} T^{\frac{1}{1+ε}})$, thus improving the dependence on $d$ for all $ε\in (0,1)$ and recovering a known optimal result for $ε= 1$. We also establish a lower bound of $Ω(d^{\frac{2ε}{1+ε}} T^{\frac{1}{1+ε}})$, which strictly improves upon the multi-armed bandit rate and highlights the hardness of heavy-tailed linear bandit problems. For finite action sets, we derive similarly improved upper and lower bounds for regret. Finally, we provide action set dependent regret upper bounds showing that for some geometries, such as $l_p$-norm balls for $p \le 1 + ε$, we can further reduce the dependence on $d$, and we can handle infinite-dimensional settings via the kernel trick, in particular establishing new regret bounds for the Matérn kernel that are the first to be sublinear for all $ε\in (0, 1]$.
-
2025-05-29 Learning to Incentivize in Repeated Principal-Agent Problems with Adversarial Agent Arrivals.
Summary: We initiate the study of a repeated principal-agent problem over a finite horizon $T$, where a principal sequentially interacts with $K\geq 2$ types of agents arriving in an adversarial order. At each round, the principal strategically chooses one of the $N$ arms to incentivize for an arriving agent of unknown type. The agent then chooses an arm based on its own utility and the provided incentive, and the principal receives a corresponding reward. The objective is to minimize regret against the best incentive in hindsight. Without prior knowledge of agent behavior, we show that the problem becomes intractable, leading to linear regret. We analyze two key settings where sublinear regret is achievable. In the first setting, the principal knows the arm each agent type would select greedily for any given incentive. Under this setting, we propose an algorithm that achieves a regret bound of $O(\min\{\sqrt{KT\log N},K\sqrt{T}\})$ and provide a matching lower bound up to a $\log K$ factor. In the second setting, an agent's response varies smoothly with the incentive and is governed by a Lipschitz constant $L\geq 1$. Under this setting, we show that there is an algorithm with a regret bound of $\tilde{O}((LN)^{1/3}T^{2/3})$ and establish a matching lower bound up to logarithmic factors. Finally, we extend our algorithmic results for both settings by allowing the principal to incentivize multiple arms simultaneously in each round.
-
2025-05-24 Improved Regret and Contextual Linear Extension for Pandora's Box and Prophet Inequality.
Summary: We study the Pandora's Box problem in an online learning setting with semi-bandit feedback. In each round, the learner sequentially pays to open up to $n$ boxes with unknown reward distributions, observes rewards upon opening, and decides when to stop. The utility of the learner is the maximum observed reward minus the cumulative cost of opened boxes, and the goal is to minimize regret defined as the gap between the cumulative expected utility and that of the optimal policy. We propose a new algorithm that achieves $\widetilde{O}(\sqrt{nT})$ regret after $T$ rounds, which improves the $\widetilde{O}(n\sqrt{T})$ bound of Agarwal et al. [2024] and matches the known lower bound up to logarithmic factors. To better capture real-life applications, we then extend our results to a natural but challenging contextual linear setting, where each box's expected reward is linear in some known but time-varying $d$-dimensional context and the noise distribution is fixed over time. We design an algorithm that learns both the linear function and the noise distributions, achieving $\widetilde{O}(nd\sqrt{T})$ regret. Finally, we show that our techniques also apply to the online Prophet Inequality problem, where the learner must decide immediately whether or not to accept a revealed reward. In both non-contextual and contextual settings, our approach achieves similar improvements and regret bounds.
-
2025-05-20 Subquadratic Algorithms and Hardness for Attention with Any Temperature.
Summary: Despite the popularity of the Transformer architecture, the standard algorithm for computing Attention suffers from quadratic time complexity in context length $n$. Alman and Song [NeurIPS 2023] showed that when the head dimension $d = Θ(\log n)$, subquadratic Attention is possible if and only if the inputs have small entries bounded by $B = o(\sqrt{\log n})$ in absolute values, under the Strong Exponential Time Hypothesis ($\mathsf{SETH}$). Equivalently, subquadratic Attention is possible if and only if the softmax is applied with high temperature for $d=Θ(\log n)$. Running times of these algorithms depend exponentially on $B$ and thus they do not lead to even a polynomial-time algorithm outside the specific range of $B$. This naturally leads to the question: when can Attention be computed efficiently without strong assumptions on temperature? Are there fast attention algorithms that scale polylogarithmically with entry size $B$? In this work, we resolve this question and characterize when fast Attention for arbitrary temperatures is possible. First, for all constant $d = O(1)$, we give the first subquadratic $\tilde{O}(n^{2 - 1/d} \cdot \mathrm{polylog}(B))$ time algorithm for Attention with large $B$. Our result holds even for matrices with large head dimension if they have low rank. In this regime, we also give a similar running time for Attention gradient computation, and therefore for the full LLM training process. Furthermore, we show that any substantial improvement on our algorithm is unlikely. In particular, we show that even when $d = 2^{Θ(\log^* n)}$, Attention requires $n^{2 - o(1)}$ time under $\mathsf{SETH}$. Finally, in the regime where $d = \mathrm{poly}(n)$, we show that the standard algorithm is optimal under popular fine-grained complexity assumptions.
-
2025-04-21 Price Stability and Improved Buyer Utility with Presentation Design: A Theoretical Study of the Amazon Buy Box.
Summary: Platforms design the form of presentation by which sellers are shown to the buyers. This design not only shapes the buyers' experience but also leads to different market equilibria or dynamics. One component in this design is through the platform's mediation of the search frictions experienced by the buyers for different sellers. We take a model of monopolistic competition and show that, on one hand, when all sellers have the same inspection costs, the market sees no stable price since the sellers always have incentives to undercut each other, and, on the other hand, the platform may stabilize the price by giving prominence to one seller chosen by a carefully designed mechanism. This calls to mind Amazon's Buy Box. We study natural mechanisms for choosing the prominent seller, characterize the range of equilibrium prices implementable by them, and find that in certain scenarios the buyers' surplus improves as the search friction increases.
-
2025-03-02 Optimal Trickle-Down Theorems for Path Complexes via C-Lorentzian Polynomials with Applications to Sampling and Log-Concave Sequences.
Summary: Let $X$ be a $d$-partite $d$-dimensional simplicial complex with parts $T_1,\dots,T_d$ and let $μ$ be a distribution on the facets of $X$. Informally, we say $(X,μ)$ is a path complex if for any $i
-
2025-02-24 S4S: Solving for a Diffusion Model Solver.
Summary: Diffusion models (DMs) create samples from a data distribution by starting from random noise and iteratively solving a reverse-time ordinary differential equation (ODE). Because each step in the iterative solution requires an expensive neural function evaluation (NFE), there has been significant interest in approximately solving these diffusion ODEs with only a few NFEs without modifying the underlying model. However, in the few NFE regime, we observe that tracking the true ODE evolution is fundamentally impossible using traditional ODE solvers. In this work, we propose a new method that learns a good solver for the DM, which we call Solving for the Solver (S4S). S4S directly optimizes a solver to obtain good generation quality by learning to match the output of a strong teacher solver. We evaluate S4S on six different pre-trained DMs, including pixel-space and latent-space DMs for both conditional and unconditional sampling. In all settings, S4S uniformly improves the sample quality relative to traditional ODE solvers. Moreover, our method is lightweight, data-free, and can be plugged in black-box on top of any discretization schedule or architecture to improve performance. Building on top of this, we also propose S4S-Alt, which optimizes both the solver and the discretization schedule. By exploiting the full design space of DM solvers, with 5 NFEs, we achieve an FID of 3.73 on CIFAR10 and 13.26 on MS-COCO, representing a $1.5\times$ improvement over previous training-free ODE methods.
-
2025-02-11 Pseudorandomness Properties of Random Reversible Circuits.
Summary: Motivated by practical concerns in cryptography, we study pseudorandomness properties of permutations on $\{0,1\}^n$ computed by random circuits made from reversible $3$-bit gates (permutations on $\{0,1\}^3$). Our main result is that a random circuit of depth $\sqrt{n} \cdot \tilde{O}(k^3)$, with each layer consisting of $Θ(n)$ random gates in a fixed two-dimensional nearest-neighbor architecture, yields approximate $k$-wise independent permutations. Our result can be seen as a particularly simple/practical block cipher construction that gives provable statistical security against attackers with access to $k$~input-output pairs within few rounds. The main technical component of our proof consists of two parts: 1. We show that the Markov chain on $k$-tuples of $n$-bit strings induced by a single random $3$-bit one-dimensional nearest-neighbor gate has spectral gap at least $1/n \cdot \tilde{O}(k)$. Then we infer that a random circuit with layers of random gates in a fixed one-dimensional gate architecture yields approximate $k$-wise independent permutations of $\{0,1\}^n$ in depth $n\cdot \tilde{O}(k^2)$ 2. We show that if the $n$ wires are layed out on a two-dimensional lattice of bits, then repeatedly alternating applications of approximate $k$-wise independent permutations of $\{0,1\}^{\sqrt n}$ to the rows and columns of the lattice yields an approximate $k$-wise independent permutation of $\{0,1\}^n$ in small depth. Our work improves on the original work of Gowers, who showed a gap of $1/\mathrm{poly}(n,k)$ for one random gate (with non-neighboring inputs); and, on subsequent work improving the gap to $Ω(1/n^2k)$ in the same setting.
-
2025-02-05 Unweighted One-Sided Code Sparsifiers and Thin Subgraphs.
Summary: For a linear code $\mathcal{C} \subseteq \mathbb{F}_2^n$ and $α\in [0,1]$, call a set $S \subseteq [n]$ an (unweighted) one-sided $α$-sparsifier of $\mathcal{C}$ if for all $c \in \mathcal{C}$, $\mathrm{wt}(c_S)\geq α\cdot \mathrm{wt}(c)$, where $c_S$ is the projection of $c$ onto the coordinates in $S$ and $\mathrm{wt}(c)$ is the Hamming weight of $c$. \\ We show that every $k$-dimensional linear code $\mathcal{C}\subseteq \mathbb{F}_2^n$ has at least $2^{n - k}$ many unweighted one-sided $1/2$-sparsifiers and hence one of size at most $n/2 + O(\sqrt{n k})$. As an application, letting $\mathcal{C} \subseteq \mathbb{F}_2^E$ denote the cut-space of a graph $G=(V, E)$, we show a lower bound of $2^{\lvert E \rvert- (\lvert V \rvert - 1)}$ on the number of $1/2$-thin subgraphs of $G$ and the existence of a $1/2$-thin subgraph with at least $\lvert E \rvert /2-O(\sqrt{\lvert E \rvert \cdot \lvert V \rvert})$ edges. In contrast to previous results on thin subgraphs, our proofs are purely "combinatorial".
-
2025-01-31 Markovian Pandora's box.
Summary: In this paper, we study the Markovian Pandora's Box Problem, where decisions are governed by both order constraints and Markovianly correlated rewards, structured within a shared directed acyclic graph. To the best of our knowledge, previous work has not incorporated Markovian dependencies in this setting. This framework is particularly relevant to applications such as data or computation driven algorithm design, where exploration of future models incurs cost. We present optimal fully adaptive strategies where the associated graph forms a forest. Under static transition, we introduce a strategy that achieves a near optimal expected payoff in multi line graphs and a 1/2 approximation in forest-structured graphs. Notably, this algorithm provides a significant speedup over the exact solution, with the improvement becoming more pronounced as the graph size increases. Our findings deepen the understanding of sequential exploration under Markovian correlations in graph-based decision-making.
-
2025-01-11 DiscQuant: A Quantization Method for Neural Networks Inspired by Discrepancy Theory.
Summary: Quantizing the weights of a neural network has two steps: (1) Finding a good low bit-complexity representation for weights (which we call the quantization grid) and (2) Rounding the original weights to values in the quantization grid. In this paper, we study the problem of rounding optimally given any quantization grid. The simplest and most commonly used way to round is Round-to-Nearest (RTN). By rounding in a data-dependent way instead, one can improve the quality of the quantized model significantly. We study the rounding problem from the lens of \emph{discrepancy theory}, which studies how well we can round a continuous solution to a discrete solution without affecting solution quality too much. We prove that given $m=\mathrm{poly}(1/ε)$ samples from the data distribution, we can round all but $O(m)$ model weights such that the expected approximation error of the quantized model on the true data distribution is $\le ε$ as long as the space of gradients of the original model is approximately low rank (which we empirically validate). Our proof, which is algorithmic, inspired a simple and practical rounding algorithm called \emph{DiscQuant}. In our experiments, we demonstrate that DiscQuant significantly improves over the prior state-of-the-art rounding method called GPTQ and the baseline RTN over a range of benchmarks on Phi3mini-3.8B and Llama3.1-8B. For example, rounding Phi3mini-3.8B to a fixed quantization grid with 3.25 bits per parameter using DiscQuant gets 64\% accuracy on the GSM8k dataset, whereas GPTQ achieves 54\% and RTN achieves 31\% (the original model achieves 84\%). We make our code available at https://github.com/jerry-chee/DiscQuant.
-
2025-01-04 A parameterized linear formulation of the integer hull.
Summary: Let $A \in \mathbb{Z}^{m \times n}$ be an integer matrix with components bounded by $Δ$ in absolute value. Cook et al.~(1986) have shown that there exists a universal matrix $B \in \mathbb{Z}^{m' \times n}$ with the following property: For each $b \in \mathbb{Z}^m$, there exists $t \in \mathbb{Z}^{m'}$ such that the integer hull of the polyhedron $P = \{ x \in \mathbb{R}^n \colon Ax \leq b\}$ is described by $P_I = \{ x \in \mathbb{R}^n \colon Bx \leq t\}$. Our \emph{main result} is that $t$ is an \emph{affine} function of $b$ as long as $b$ is from a fixed equivalence class of the lattice $D \cdot \mathbb{Z}^m$. Here $D \in \mathbb{N}$ is a number that depends on $n$ and $Δ$ only. Furthermore, $D$ as well as the matrix $B$ can be computed in time depending on $Δ$ and $n$ only. An application of this result is the solution of an open problem posed by Cslovjecsek et al.~(SODA 2024) concerning the complexity of \emph{2-stage-stochastic integer programming} problems. The main tool of our proof is the classical theory of \emph{Chvátal-Gomory cutting planes} and the \emph{elementary closure} of rational polyhedra.