papers

Publications (61)

math.OC2019

Proximal algorithms for constrained composite optimization, with applications to solving low-rank SDPs

Yu Bai, John Duchi, Song Mei

We study a family of (potentially non-convex) constrained optimization problems with convex composite structure. Through a novel analysis of non-smooth geometry, we show that proxi…

cs.LG2022

When Can We Learn General-Sum Markov Games with a Large Number of Players Sample-Efficiently?

Ziang Song, Song Mei, Yu Bai

Multi-agent reinforcement learning has made substantial empirical progresses in solving games with a large number of players. However, theoretically, the best known sample complexi…

cs.LG2022

Efficient and Differentiable Conformal Prediction with General Function Classes

Yu Bai, Song Mei, Huan Wang +2

Quantifying the data uncertainty in learning tasks is often done by learning a prediction interval or prediction set of the label given the input. Two commonly desired properties f…

cs.LG2024

Active-Dormant Attention Heads: Mechanistically Demystifying Extreme-Token Phenomena in LLMs

Tianyu Guo, Druv Pai, Yu Bai +3

Practitioners have consistently observed three puzzling phenomena in transformer-based large language models (LLMs): attention sinks, value-state drains, and residual-state peaks,…

stat.ME2023

Uncertainty Intervals for Prediction Errors in Time Series Forecasting

Hui Xu, Song Mei, Stephen Bates +2

Inference for prediction errors is critical in time series forecasting pipelines. However, providing statistically meaningful uncertainty intervals for prediction errors remains re…

cs.CL2025

Improving LLM Safety Alignment with Dual-Objective Optimization

Xuandong Zhao, Will Cai, Tianneng Shi +4

Existing training-time safety alignment techniques for large language models (LLMs) remain vulnerable to jailbreak attacks. Direct preference optimization (DPO), a widely deployed…

cs.LG2024

Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning

Ruiqi Zhang, Licong Lin, Yu Bai +1

Large Language Models (LLMs) often memorize sensitive, private, or copyrighted data during pre-training. LLM unlearning aims to eliminate the influence of undesirable data from the…

cs.LG2024

An Overview of Diffusion Models: Applications, Guided Generation, Statistical Rates and Optimization

Minshuo Chen, Song Mei, Jianqing Fan +1

Diffusion models, a powerful and universal generative AI technology, have achieved tremendous success in computer vision, audio, reinforcement learning, and computational biology.…

cs.CL2025

How Do LLMs Perform Two-Hop Reasoning in Context?

Tianyu Guo, Hanlin Zhu, Ruiqi Zhang +4

``Socrates is human. All humans are mortal. Therefore, Socrates is mortal.'' This form of argument illustrates a typical pattern of two-hop reasoning. Formally, two-hop reasoning r…

cs.AI2025

GSM-Agent: Understanding Agentic Reasoning Using Controllable Environments

Hanlin Zhu, Tianyu Guo, Song Mei +4

As LLMs are increasingly deployed as agents, agentic reasoning - the ability to combine tool use, especially search, and reasoning - becomes a critical skill. However, it is hard t…

stat.ML2017

The Landscape of Empirical Risk for Non-convex Losses

Song Mei, Yu Bai, Andrea Montanari

Most high-dimensional estimation and prediction methods propose to minimize a cost function (empirical risk) that is written as a sum of losses associated to each data point. In th…

cs.LG2024

Transformers as Decision Makers: Provable In-Context Reinforcement Learning via Supervised Pretraining

Licong Lin, Yu Bai, Song Mei

Large transformer models pretrained on offline reinforcement learning datasets have demonstrated remarkable in-context reinforcement learning (ICRL) capabilities, where they can ma…

math.ST2021

Generalization error of random features and kernel methods: hypercontractivity and kernel matrix concentration

Song Mei, Theodor Misiakiewicz, Andrea Montanari

Consider the classical supervised learning problem: we are given data , , with a response and a covariate…

cs.LG2024

U-Nets as Belief Propagation: Efficient Classification, Denoising, and Diffusion in Generative Hierarchical Models

Song Mei

U-Nets are among the most widely used architectures in computer vision, renowned for their exceptional performance in applications such as image segmentation, denoising, and diffus…

cs.LG2026

SPEED-RL: Faster Training of Reasoning Models via Online Curriculum Learning

Ruiqi Zhang, Daman Arora, Song Mei +1

Training large language models with reinforcement learning (RL) against verifiable rewards significantly enhances their reasoning abilities, yet remains computationally expensive d…

stat.ML2024

Large Stepsize Gradient Descent for Non-Homogeneous Two-Layer Networks: Margin Improvement and Fast Optimization

Yuhang Cai, Jingfeng Wu, Song Mei +2

The typical training of neural networks using large stepsize gradient descent (GD) under the logistic loss often involves two distinct phases, where the empirical risk oscillates i…

cs.LG2023

Lower Bounds for Learning in Revealing POMDPs

Fan Chen, Huan Wang, Caiming Xiong +2

This paper studies the fundamental limits of reinforcement learning (RL) in the challenging \emph{partially observable} setting. While it is well-established that learning in Parti…

math.ST2018

The landscape of the spiked tensor model

Gerard Ben Arous, Song Mei, Andrea Montanari +1

We consider the problem of estimating a large rank-one tensor , in Gaussian noise. Earlier work characterized…

cs.LG2025

A Statistical Theory of Contrastive Pre-training and Multimodal Generative AI

Kazusato Oko, Licong Lin, Yuhang Cai +1

Multi-modal generative AI systems, such as those combining vision and language, rely on contrastive pre-training to learn representations across different modalities. While their p…

cs.LG2023

Near-Optimal Learning of Extensive-Form Games with Imperfect Information

Yu Bai, Chi Jin, Song Mei +1

This paper resolves the open question of designing near-optimal algorithms for learning imperfect-information extensive-form games from bandit feedback. We present the first line o…

stat.ML2018

A Mean Field View of the Landscape of Two-Layers Neural Networks

Song Mei, Andrea Montanari, Phan-Minh Nguyen

Multi-layer neural networks are among the most powerful models in machine learning, yet the fundamental reasons for this success defy mathematical understanding. Learning a neural…

cs.CL2026

DiscoLoop: Looping Discrete Embeddings and Continuous Hidden States for Multi-hop Reasoning

Hengyu Fu, Tianyu Guo, Zixuan Wang +5

Large language models achieve strong performance on many reasoning tasks when allowed to externalize intermediate steps as Chain-of-Thought (CoT). However, many questions require t…

math.ST2023

Local convexity of the TAP free energy and AMP convergence for Z2-synchronization

Michael Celentano, Zhou Fan, Song Mei

We study mean-field variational Bayesian inference using the TAP approach, for Z2-synchronization as a prototypical example of a high-dimensional Bayesian model. We show that for a…

stat.ML2021

Learning with invariances in random features and kernel models

Song Mei, Theodor Misiakiewicz, Andrea Montanari

A number of machine learning tasks entail a high degree of invariance: the data distribution does not change if we act on the data with a certain group of transformations. For inst…

quant-ph2022

Performance and limitations of the QAOA at constant levels on large sparse hypergraphs and spin glass models

Joao Basso, David Gamarnik, Song Mei +1

The Quantum Approximate Optimization Algorithm (QAOA) is a general purpose quantum algorithm designed for combinatorial optimization. We analyze its expected performance and prove…

cs.LG2022

Efficient Phi-Regret Minimization in Extensive-Form Games via Online Mirror Descent

Yu Bai, Chi Jin, Song Mei +2

A conceptually appealing approach for learning Extensive-Form Games (EFGs) is to convert them to Normal-Form Games (NFGs). This approach enables us to directly translate state-of-t…

stat.ML2025

An Overview of Large Language Models for Statisticians

Wenlong Ji, Weizhe Yuan, Emily Getzen +7

Large Language Models (LLMs) have emerged as transformative tools in artificial intelligence (AI), exhibiting remarkable capabilities across diverse tasks such as text generation,…

cs.LG2021

Don't Just Blame Over-parametrization for Over-confidence: Theoretical Analysis of Calibration in Binary Classification

Yu Bai, Song Mei, Huan Wang +1

Modern machine learning models with high accuracy are often miscalibrated -- the predicted top probability does not reflect the actual accuracy, and tends to be over-confident. It…

cs.LG2025

Implicit Bias of Gradient Descent for Non-Homogeneous Deep Networks

Yuhang Cai, Kangjie Zhou, Jingfeng Wu +3

We establish the asymptotic implicit bias of gradient descent (GD) for generic non-homogeneous deep networks under exponential loss. Specifically, we characterize three key propert…

cs.LG2023

Deep Networks as Denoising Algorithms: Sample-Efficient Learning of Diffusion Models in High-Dimensional Graphical Models

Song Mei, Yuchen Wu

We investigate the approximation efficiency of score functions by deep neural networks in diffusion-based generative modeling. While existing approximation theories utilize the smo…

cond-mat.soft2015

On a molecular based Q-tensor model for liquid crystals with density variations

Song Mei, Pingwen Zhang

In this article, we study the new Q-tensor model previously derived from Onsager's molecular theory by Han \textit{et al.} [Arch. Rational Mech. Anal., 215.3 (2014), pp. 741-809] f…

quant-ph2024

Statistical Estimation in the Spiked Tensor Model via the Quantum Approximate Optimization Algorithm

Leo Zhou, Joao Basso, Song Mei

The quantum approximate optimization algorithm (QAOA) is a general-purpose algorithm for combinatorial optimization. In this paper, we analyze the performance of the QAOA on a stat…

cs.LG2022

Partially Observable RL with B-Stability: Unified Structural Condition and Sharp Sample-Efficient Algorithms

Fan Chen, Yu Bai, Song Mei

Partial Observability -- where agents can only observe partial information about the true underlying state of the system -- is ubiquitous in real-world applications of Reinforcemen…

math.ST2020

The generalization error of random features regression: Precise asymptotics and double descent curve

Song Mei, Andrea Montanari

Deep learning methods operate in regimes that defy the traditional statistical mindset. Neural network architectures often contain more parameters than training samples, and are so…

stat.ML2025

A Statistical Theory of Contrastive Learning via Approximate Sufficient Statistics

Licong Lin, Song Mei

Contrastive learning -- a modern approach to extract useful representations from unlabeled data by training models to distinguish similar samples from dissimilar ones -- has driven…

cs.CL2025

Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning

Chongyu Fan, Jiancheng Liu, Licong Lin +4

This work studies the problem of large language model (LLM) unlearning, aiming to remove unwanted data influences (e.g., copyrighted or harmful content) while preserving model util…

cs.LG2025

OVERT: A Benchmark for Over-Refusal Evaluation on Text-to-Image Models

Ziheng Cheng, Yixiao Huang, Hui Xu +4

Text-to-Image (T2I) models have achieved remarkable success in generating visual content from text inputs. Although multiple safety alignment strategies have been proposed to preve…

cs.LG2021

Exact Gap between Generalization Error and Uniform Convergence in Random Feature Models

Zitong Yang, Yu Bai, Song Mei

Recent work showed that there could be a large gap between the classical uniform convergence bound and the actual test error of zero-training-error predictors (interpolators) such…

math.ST2023

Mean-field variational inference with the TAP free energy: Geometric and statistical properties in linear models

Michael Celentano, Zhou Fan, Licong Lin +1

We study mean-field variational inference in a Bayesian linear model when the sample size n is comparable to the dimension p. In high dimensions, the common approach of minimizing…

cs.LG2022

Sample-Efficient Learning of Correlated Equilibria in Extensive-Form Games

Ziang Song, Song Mei, Yu Bai

Imperfect-Information Extensive-Form Games (IIEFGs) is a prevalent model for real-world games involving imperfect information and sequential plays. The Extensive-Form Correlated Eq…

cs.CL2026

OpenAI GPT-5 System Card

Aaditya Singh, Adam Fry, Adam Perelman +483

This is the system card published alongside the OpenAI GPT-5 launch, August 2025. GPT-5 is a unified system with a smart and fast model that answers most questions, a deeper reason…

math.ST2023

Near-optimal multiple testing in Bayesian linear models with finite-sample FDR control

Taejoo Ahn, Licong Lin, Song Mei

In high dimensional variable selection problems, statisticians often seek to design multiple testing procedures that control the False Discovery Rate (FDR), while concurrently iden…

stat.ML2021

When Do Neural Networks Outperform Kernel Methods?

Behrooz Ghorbani, Song Mei, Theodor Misiakiewicz +1

For a certain scaling of the initialization of stochastic gradient descent (SGD), wide neural networks (NN) have been shown to be well approximated by reproducing kernel Hilbert sp…

cs.CL2026

Generalization or Hallucination? Understanding Out-of-Context Reasoning in Transformers

Yixiao Huang, Hanlin Zhu, Tianyu Guo +5

Large language models (LLMs) can acquire new knowledge through fine-tuning, but this process exhibits a puzzling duality: models can generalize remarkably from new facts, yet are a…

cs.LG2023

Transformers as Statisticians: Provable In-Context Learning with In-Context Algorithm Selection

Yu Bai, Fan Chen, Huan Wang +2

Neural sequence models based on the transformer architecture have demonstrated remarkable \emph{in-context learning} (ICL) abilities, where they can perform new tasks when prompted…

cs.LG2023

How Do Transformers Learn In-Context Beyond Simple Functions? A Case Study on Learning with Representations

Tianyu Guo, Wei Hu, Song Mei +4

While large language models based on the transformer architecture have demonstrated remarkable in-context learning (ICL) capabilities, understandings of such capabilities are still…

cs.LG2021

Understanding the Under-Coverage Bias in Uncertainty Estimation

Yu Bai, Song Mei, Huan Wang +1

Estimating the data uncertainty in regression tasks is often done by learning a quantile function or a prediction interval of the true label conditioned on the input. It is frequen…

math.ST2020

Linearized two-layers neural networks in high dimension

Behrooz Ghorbani, Song Mei, Theodor Misiakiewicz +1

We consider the problem of learning an unknown function on the -dimensional sphere with respect to the square loss, given i.i.d. samples $\{(y_i,{\boldsymbol x}_i)\}…

stat.ML2019

Mean-field theory of two-layers neural networks: dimension-free bounds and kernel limit

Song Mei, Theodor Misiakiewicz, Andrea Montanari

We consider learning two layer neural networks using stochastic gradient descent. The mean-field description of this learning dynamics approximates the evolution of the network wei…

math.OC2017

Solving SDPs for synchronization and MaxCut problems via the Grothendieck inequality

Song Mei, Theodor Misiakiewicz, Andrea Montanari +1

A number of statistical estimation problems can be addressed by semidefinite programs (SDP). While SDPs are solvable in polynomial time using interior point methods, in practice ge…

cs.LG2024

Unified Algorithms for RL with Decision-Estimation Coefficients: PAC, Reward-Free, Preference-Based Learning, and Beyond

Fan Chen, Song Mei, Yu Bai

Modern Reinforcement Learning (RL) is more than just learning the optimal policy; Alternative learning goals such as exploring the environment, estimating the underlying model, and…

cs.SI2022

A Novel Location Free Link Prediction in Multiplex Social Networks

Song Mei, Cong Zhen

In recent decades, the emergence of social networks has enabled internet service providers (e.g., Facebook, Twitter and Uber) to achieve great commercial success. Link prediction i…

cs.CR2022

Plausible deniability for privacy-preserving data synthesis

Song Mei, Zhiqiang Ye

In the field of privacy protection, publishing complete data (especially high-dimensional data sets) is one of the most challenging problems. The common encryption technology can n…

stat.ML2021

The Three Stages of Learning Dynamics in High-Dimensional Kernel Methods

Nikhil Ghosh, Song Mei, Bin Yu

To understand how deep learning works, it is crucial to understand the training dynamics of neural networks. Several interesting hypotheses about these dynamics have been made base…

math.PR2020

TAP free energy, spin glasses, and variational inference

Zhou Fan, Song Mei, Andrea Montanari

We consider the Sherrington-Kirkpatrick model of spin glasses with ferromagnetically biased couplings. For a specific choice of the couplings mean, the resulting Gibbs measure is e…

cs.LG2026

Transformers Provably Learn to Internalize Chain-of-Thought

Yixiao Huang, Hanlin Zhu, Zixuan Wang +4

Chain-of-Thought (CoT) prompting substantially improves the sample efficiency of transformers, reducing the complexity of tasks like parity learning from exponential to polynomial…

cs.LG2023

What can a Single Attention Layer Learn? A Study Through the Random Features Lens

Hengyu Fu, Tianyu Guo, Yu Bai +1

Attention layers -- which map a sequence of inputs to a sequence of outputs -- are core building blocks of the Transformer architecture which has achieved significant breakthroughs…

stat.ML2022

Learning with convolution and pooling operations in kernel methods

Theodor Misiakiewicz, Song Mei

Recent empirical work has shown that hierarchical convolutional kernels inspired by convolutional neural networks (CNNs) significantly improve the performance of kernel methods in…

stat.ML2019

Limitations of Lazy Training of Two-layers Neural Networks

Behrooz Ghorbani, Song Mei, Theodor Misiakiewicz +1

We study the supervised learning problem under either of the following two models: (1) Feature vectors are -dimensional Gaussians and responses are $y_i = f_…

cs.CL2025

gpt-oss-120b & gpt-oss-20b Model Card

OpenAI, :, Sandhini Agarwal +124

We present gpt-oss-120b and gpt-oss-20b, two open-weight reasoning models that push the frontier of accuracy and inference cost. The models use an efficient mixture-of-expert trans…

math.OC2019

Analysis of Sequential Quadratic Programming through the Lens of Riemannian Optimization

Yu Bai, Song Mei

We prove that a "first-order" Sequential Quadratic Programming (SQP) algorithm for equality constrained optimization has local linear convergence with rate , where $Î…