Searcharxiv⌕ Search

arXiv subjects

Gregory Valiant

Publications and source records attributed to Gregory Valiant.

At least 19 recordsLinked to original sources
arXiv2026

Thompson Sampling Is 2-Competitive for Mistakes

We consider Bayesian bandit models and prove that Thompson sampling makes at most twice the expected number of mistakes (selections of a suboptimal arm) as any other policy. Our analysis applies as long as the latent arm processes are independent and each arm evolves only when played. For stochastic bandits with best arm defined via mean reward, this confirms a conjecture of Guha and Munagala from 2014, where the factor $2$ is already best possible. The result holds under any nonincreasing sequence of round weights, including fixed horizon and geometric discounting.

stat.ML↗
arXiv2026

Short proofs in combinatorics, probability and number theory II

We give a quintet of proofs resulting from questions posed by Erd\H{o}s. These questions concern ordinary lines in planar point sets, sequences with uniformly small exponential sums, $K_4$-free $4$-critical graphs with few chords in any cycle, a counterexample to a "fewnomial" version of the Erd\H{o}s--Tur\'{a}n discrepancy bound, and a finiteness theorem for integers $n$ such that $n-a k^2$ is prime for all $k\leq \sqrt{n/a}$ coprime to $n$ (for fixed $a\in\mathbb Z_+$). Each proof is due to an internal model at OpenAI.

math.CO↗
arXiv2026

Short proofs in combinatorics and number theory

We give a triplet of short proofs, each of which answers a question raised by Erd\H{o}s. The first concerns the small prime factors of $\binom{n}{k}$, the second concerns whether an additive basis $A$ can always be split into pieces $A_1$ and $A_2$ such that each of $A_i + A_i$ has bounded gaps, and the final concerns whether $\{\alpha p\}$ is "well-distributed" in the sense introduced by Hlawka and Petersen. In each case, the proof is due entirely to an internal model at OpenAI.

math.CO↗
arXiv2026

On infinite sets with no $3$ on a line

We give a construction of an infinite set of points $A$ in $\mathbb{R}^2$ such that any subset $P\subseteq A$ has a constant density subset $P'$ with no three points collinear and yet $A$ cannot be separated into finitely many subsets such that each subset has no three points collinear. This provides a new proof of a question of Erd\H{o}s, Ne\v{s}et\v{r}il, and R\"{o}dl. The construction was generated by an internal model at OpenAI.

math.CO↗
arXiv2025

On the Entropy Calibration of Language Models

We study the problem of entropy calibration, which asks whether a language model's entropy over generations matches its log loss on human text. Past work found that models are miscalibrated, with entropy per step increasing as generations grow longer, due to error accumulation. To calibrate the model and improve text quality, it has become standard practice to truncate the distribution, but this approach reduces output diversity, which we would like to avoid. Therefore, in this paper, we ask: does miscalibration improve automatically with scale, and if not, is it theoretically possible to calibrate without tradeoffs? To build intuition, we first study a simplified theoretical setting to characterize the scaling behavior of miscalibration with respect to dataset size. We find that the rate of scaling depends on the power law exponent of the data distribution -- in particular, for a power law exponent close to 1, the scaling exponent is close to 0, meaning that miscalibration improves very slowly with scale. Next, we measure miscalibration empirically in language models ranging from 0.5B to 70B parameters. We find that the observed scaling behavior is similar to what is predicted theoretically: our fitted scaling exponents for text are close to 0, meaning that larger models accumulate error at a similar rate as smaller ones. This scaling (or, lack thereof) provides one explanation for why we sample from larger models with similar amounts of truncation as smaller models, even though the larger models are of higher quality. However, truncation is not a satisfying solution because it comes at the cost of increased log loss. In theory, is it even possible to reduce entropy while preserving log loss? We prove that it is possible, if we assume access to a black box which can fit models to predict the future entropy of text.

cs.CL↗
arXiv2025

A Simple Geometric Proof of the Optimality of the Sequential Probability Ratio Test for Symmetric Bernoulli Hypotheses

This paper revisits the classical problem of determining the bias of a weighted coin, where the bias is known to be either $p = 1/2 + \varepsilon$ or $p = 1/2 - \varepsilon$, while minimizing the expected number of coin tosses and the error probability. The optimal strategy for this problem is given by Wald's Sequential Probability Ratio Test (SPRT), which compares the log-likelihood ratio against fixed thresholds to determine a stopping time. Classical proofs of this result typically rely on analytical, continuous, and non-constructive arguments. In this paper, we present a discrete, self-contained proof of the optimality of the SPRT for this problem. We model the problem as a biased random walk on the two-dimensional (heads, tails) integer lattice, and model strategies as marked stopping times on this lattice. Our proof takes a straightforward greedy approach, showing how any arbitrary strategy may be transformed into the optimal, parallel-line "difference policy" corresponding to the SPRT, via a sequence of local perturbations that improve a Bayes risk objective.

math.ST↗
arXiv2025

Attainability of Two-Point Testing Rates for Finite-Sample Location Estimation

Le Cam's two-point testing method yields perhaps the simplest lower bound for estimating the mean of a distribution: roughly, if it is impossible to well-distinguish a distribution centered at $\mu$ from the same distribution centered at $\mu+\Delta$, then it is impossible to estimate the mean by better than $\Delta/2$. It is setting-dependent whether or not a nearly matching upper bound is attainable. We study the conditions under which the two-point testing lower bound can be attained for univariate mean estimation; both in the setting of location estimation (where the distribution is known up to translation) and adaptive location estimation (unknown distribution). Roughly, we will say an estimate nearly attains the two-point testing lower bound if it incurs error that is at most polylogarithmically larger than the Hellinger modulus of continuity for $\tilde{\Omega}(n)$ samples. Adaptive location estimation is particularly interesting as some distributions admit much better guarantees than sub-Gaussian rates (e.g. $\operatorname{Unif}(\mu-1,\mu+1)$ permits error $\Theta(\frac{1}{n})$, while the sub-Gaussian rate is $\Theta(\frac{1}{\sqrt{n}})$), yet it is not obvious whether these rates may be adaptively attained by one unified approach. Our main result designs an algorithm that nearly attains the two-point testing rate for mixtures of symmetric, log-concave distributions with a common mean. Moreover, this algorithm runs in near-linear time and is parameter-free. In contrast, we show the two-point testing rate is not nearly attainable even for symmetric, unimodal distributions. We complement this with results for location estimation, showing the two-point testing rate is nearly attainable for unimodal distributions, but unattainable for symmetric distributions.

math.ST↗
arXiv2024

A Generalized Trace Reconstruction Problem: Recovering a String of Probabilities

We introduce the following natural generalization of trace reconstruction, parameterized by a deletion probability $\delta \in (0,1)$ and length $n$: There is a length $n$ string of probabilities, $S=p_1,\ldots,p_n,$ and each "trace" is obtained by 1) sampling a length $n$ binary string whose $i$th coordinate is independently set to 1 with probability $p_i$ and 0 otherwise, and then 2) deleting each of the binary values independently with probability $\delta$, and returning the corresponding binary string of length $\le n$. The goal is to recover an estimate of $S$ from a set of independently drawn traces. In the case that all $p_i \in \{0,1\}$ this is the standard trace reconstruction problem. We show two complementary results. First, for worst-case strings $S$ and any deletion probability at least order $1/\sqrt{n}$, no algorithm can approximate $S$ to constant $\ell_\infty$ distance or $\ell_1$ distance $o(\sqrt n)$ using fewer than $2^{\Omega(\sqrt{n})}$ traces. Second -- as in the case for standard trace reconstruction -- reconstruction is easy for random $S$: for any sufficiently small constant deletion probability, and any $\epsilon>0$, drawing each $p_i$ independently from the uniform distribution over $[0,1]$, with high probability $S$ can be recovered to $\ell_1$ error $\epsilon$ using $\mathrm{poly}(n,1/\epsilon)$ traces and computation time. We show indistinguishability in our lower bound by regarding a complicated alternating sum (comparing two distributions) as the Fourier transformation of some function evaluated at $\pm \pi,$ and then showing that the Fourier transform decays rapidly away from zero by analyzing its moment generating function.

cs.DS↗
arXiv2024

Discovering Data Structures: Nearest Neighbor Search and Beyond

We propose a general framework for end-to-end learning of data structures. Our framework adapts to the underlying data distribution and provides fine-grained control over query and space complexity. Crucially, the data structure is learned from scratch, and does not require careful initialization or seeding with candidate data structures/algorithms. We first apply this framework to the problem of nearest neighbor search. In several settings, we are able to reverse-engineer the learned data structures and query algorithms. For 1D nearest neighbor search, the model discovers optimal distribution (in)dependent algorithms such as binary search and variants of interpolation search. In higher dimensions, the model learns solutions that resemble k-d trees in some regimes, while in others, they have elements of locality-sensitive hashing. The model can also learn useful representations of high-dimensional data and exploit them to design effective data structures. We also adapt our framework to the problem of estimating frequencies over a data stream, and believe it could also be a powerful discovery tool for new problems.

cs.LG↗
arXiv2024

Adaptive and oblivious statistical adversaries are equivalent

We resolve a fundamental question about the ability to perform a statistical task, such as learning, when an adversary corrupts the sample. Such adversaries are specified by the types of corruption they can make and their level of knowledge about the sample. The latter distinguishes between sample-adaptive adversaries which know the contents of the sample when choosing the corruption, and sample-oblivious adversaries, which do not. We prove that for all types of corruptions, sample-adaptive and sample-oblivious adversaries are \emph{equivalent} up to polynomial factors in the sample size. This resolves the main open question introduced by [BLMT22] and further explored in [CHL+23]. Specifically, consider any algorithm $A$ that solves a statistical task even when a sample-oblivious adversary corrupts its input. We show that there is an algorithm $A'$ that solves the same task when the corresponding sample-adaptive adversary corrupts its input. The construction of $A'$ is simple and maintains the computational efficiency of $A$: It requests a polynomially larger sample than $A$ uses and then runs $A$ on a uniformly random subsample.

cs.LG↗
arXiv2023

Near-Optimal Mean Estimation with Unknown, Heteroskedastic Variances

Given data drawn from a collection of Gaussian variables with a common mean but different and unknown variances, what is the best algorithm for estimating their common mean? We present an intuitive and efficient algorithm for this task. As different closed-form guarantees can be hard to compare, the Subset-of-Signals model serves as a benchmark for heteroskedastic mean estimation: given $n$ Gaussian variables with an unknown subset of $m$ variables having variance bounded by 1, what is the optimal estimation error as a function of $n$ and $m$? Our algorithm resolves this open question up to logarithmic factors, improving upon the previous best known estimation error by polynomial factors when $m = n^c$ for all $0<c<1$. Of particular note, we obtain error $o(1)$ with $m = \tilde{O}(n^{1/4})$ variance-bounded samples, whereas previous work required $m = \tilde{\Omega}(n^{1/2})$. Finally, we show that in the multi-dimensional setting, even for $d=2$, our techniques enable rates comparable to knowing the variance of each sample.

math.ST↗
arXiv2023

Matrix Multiplication in Quadratic Time and Energy? Towards a Fine-Grained Energy-Centric Church-Turing Thesis

We describe two algorithms for multiplying n x n matrices using time and energy n^2 polylog(n) under basic models of classical physics. The first algorithm is for multiplying integer-valued matrices, and the second, quite different algorithm, is for Boolean matrix multiplication. We hope this work inspires a deeper consideration of physically plausible/realizable models of computing that might allow for algorithms which improve upon the runtimes and energy usages suggested by the parallel RAM model in which each operation requires one unit of time and one unit of energy.

cs.CC↗
arXiv2023

Testing with Non-identically Distributed Samples

We examine the extent to which sublinear-sample property testing and estimation apply to settings where samples are independently but not identically distributed. Specifically, we consider the following distributional property testing framework: Suppose there is a set of distributions over a discrete support of size $k$, $p_1, p_2,\ldots,p_T$, and we obtain $c$ independent draws from each distribution. Suppose the goal is to learn or test a property of the average distribution, $p_{avg}$. This setup models a number of important practical settings where the individual distributions correspond to heterogeneous entities -- either individuals, chronologically distinct time periods, spatially separated data sources, etc. From a learning standpoint, even with $c=1$ samples from each distribution, $\Theta(k/\varepsilon^2)$ samples are necessary and sufficient to learn $p_{avg}$ to within error $\varepsilon$ in $\ell_1$ distance. To test uniformity or identity -- distinguishing the case that $p_{avg}$ is equal to some reference distribution, versus has $\ell_1$ distance at least $\varepsilon$ from the reference distribution, we show that a linear number of samples in $k$ is necessary given $c=1$ samples from each distribution. In contrast, for $c \ge 2$, we recover the usual sublinear sample testing guarantees of the i.i.d.\ setting: we show that $O(\sqrt{k}/\varepsilon^2 + 1/\varepsilon^4)$ total samples are sufficient, matching the optimal sample complexity in the i.i.d.\ case in the regime where $\varepsilon \ge k^{-1/4}$. Additionally, we show that in the $c=2$ case, there is a constant $\rho > 0$ such that even in the linear regime with $\rho k$ samples, no tester that considers the multiset of samples (ignoring which samples were drawn from the same $p_i$) can perform uniformity testing. We also extend our techniques to the problem of testing "closeness" of two distributions.

cs.DS↗
arXiv2023

One-sided Matrix Completion from Two Observations Per Row

Given only a few observed entries from a low-rank matrix $X$, matrix completion is the problem of imputing the missing entries, and it formalizes a wide range of real-world settings that involve estimating missing data. However, when there are too few observed entries to complete the matrix, what other aspects of the underlying matrix can be reliably recovered? We study one such problem setting, that of "one-sided" matrix completion, where our goal is to recover the right singular vectors of $X$, even in the regime where recovering the left singular vectors is impossible, which arises when there are more rows than columns and very few observations. We propose a natural algorithm that involves imputing the missing values of the matrix $X^TX$ and show that even with only two observations per row in $X$, we can provably recover $X^TX$ as long as we have at least $\Omega(r^2 d \log d)$ rows, where $r$ is the rank and $d$ is the number of columns. We evaluate our algorithm on one-sided recovery of synthetic data and low-coverage genome sequencing. In these settings, our algorithm substantially outperforms standard matrix completion and a variety of direct factorization methods.

cs.LG↗
arXiv2023

Lexinvariant Language Models

Token embeddings, a mapping from discrete lexical symbols to continuous vectors, are at the heart of any language model (LM). However, lexical symbol meanings can also be determined and even redefined by their structural role in a long context. In this paper, we ask: is it possible for a language model to be performant without \emph{any} fixed token embeddings? Such a language model would have to rely entirely on the co-occurence and repetition of tokens in the context rather than the \textit{a priori} identity of any token. To answer this, we study \textit{lexinvariant}language models that are invariant to lexical symbols and therefore do not need fixed token embeddings in practice. First, we prove that we can construct a lexinvariant LM to converge to the true language model at a uniform rate that is polynomial in terms of the context length, with a constant factor that is sublinear in the vocabulary size. Second, to build a lexinvariant LM, we simply encode tokens using random Gaussian vectors, such that each token maps to the same representation within each sequence but different representations across sequences. Empirically, we demonstrate that it can indeed attain perplexity comparable to that of a standard language model, given a sufficiently long context. We further explore two properties of the lexinvariant language models: First, given text generated from a substitution cipher of English, it implicitly implements Bayesian in-context deciphering and infers the mapping to the underlying real tokens with high accuracy. Second, it has on average 4X better accuracy over synthetic in-context reasoning tasks. Finally, we discuss regularizing standard language models towards lexinvariance and potential practical applications.

cs.CL↗
arXiv2022

Online Pen Testing

We study a "pen testing" problem, in which we are given $n$ pens with unknown amounts of ink $X_1, X_2, \ldots, X_n$, and we want to choose a pen with the maximum amount of remaining ink in it. The challenge is that we cannot access each $X_i$ directly; we only get to write with the $i$-th pen until either a certain amount of ink is used, or the pen runs out of ink. In both cases, this testing reduces the remaining ink in the pen and thus the utility of selecting it. Despite this significant lack of information, we show that it is possible to approximately maximize our utility up to an $O(\log n)$ factor. Formally, we consider two different setups: the "prophet" setting, in which each $X_i$ is independently drawn from some distribution $\mathcal{D}_i$, and the "secretary" setting, in which $(X_i)_{i=1}^n$ is a random permutation of arbitrary $a_1, a_2, \ldots, a_n$. We derive the optimal competitive ratios in both settings up to constant factors. Our algorithms are surprisingly robust: (1) In the prophet setting, we only require one sample from each $\mathcal{D}_i$, rather than a full description of the distribution; (2) In the secretary setting, the algorithm also succeeds under an arbitrary permutation, if an estimate of the maximum $a_i$ is given. Our techniques include a non-trivial online sampling scheme from a sequence with an unknown length, as well as the construction of a hard, non-uniform distribution over permutations. Both might be of independent interest. We also highlight some immediate open problems and discuss several directions for future research.

cs.DS↗
arXiv2022

What Can Transformers Learn In-Context? A Case Study of Simple Function Classes

In-context learning refers to the ability of a model to condition on a prompt sequence consisting of in-context examples (input-output pairs corresponding to some task) along with a new query input, and generate the corresponding output. Crucially, in-context learning happens only at inference time without any parameter updates to the model. While large language models such as GPT-3 exhibit some ability to perform in-context learning, it is unclear what the relationship is between tasks on which this succeeds and what is present in the training data. To make progress towards understanding in-context learning, we consider the well-defined problem of training a model to in-context learn a function class (e.g., linear functions): that is, given data derived from some functions in the class, can we train a model to in-context learn "most" functions from this class? We show empirically that standard Transformers can be trained from scratch to perform in-context learning of linear functions -- that is, the trained model is able to learn unseen linear functions from in-context examples with performance comparable to the optimal least squares estimator. In fact, in-context learning is possible even under two forms of distribution shift: (i) between the training data of the model and inference-time prompts, and (ii) between the in-context examples and the query input during inference. We also show that we can train Transformers to in-context learn more complex function classes -- namely sparse linear functions, two-layer neural networks, and decision trees -- with performance that matches or exceeds task-specific learning algorithms. Our code and models are available at https://github.com/dtsip/in-context-learning .

cs.CL↗
arXiv2022

From Sand to Flour: The Next Leap in Granular Computing with NanoSort

The granularity of distributed computing is limited by communication time: there is no point in farming out smaller and smaller tasks if the communication overhead dominates the decrease in processing time due to the added parallelism. In this work, we leverage the low communication latency of a new NIC/CPU hardware design, the nanoPU, to explore a new extreme of granularity in distributed computation, where a problem is partitioned into tens of thousands of nanosecond-scale tasks. To evaluate the feasibility and practicality of extremely fine-grained computing, we built NanoSort, a distributed sorting algorithm running on the nanoPU. Using cycle-accurate FireSim simulations of 65,536 nanoPU cores, we show that NanoSort can sort 1M keys in 68$\mu$s, an order of magnitude faster than MilliSort, the current state-of-the-art.

cs.DC↗