Taiji Suzuki

In-Context Learning Is Provably Bayesian Inference: A Generalization Theory for Meta-Learning

Inference-time Alignment with Rewards in Anisotropic Besov Spaces: Superiority of Neural Networks over Linear Estimators

Consistency Is Not Always Correct: Towards Understanding the Role of Exploration in Post-Training Reasoning

Mamba Can Learn Low-Dimensional Targets In-Context via Test-Time Feature Learning

Transformers as Measure-Theoretic Associative Memory: A Statistical Perspective and Minimax Optimality

Test time training enhances in-context learning of nonlinear functions

Provable Benefit of Curriculum in Transformer Tree-Reasoning Post-Training

Transformers Provably Solve Parity Efficiently with Chain of Thought

Hessian-guided Perturbed Wasserstein Gradient Flows for Escaping Saddle Points

Weighted Point Set Embedding for Multimodal Contrastive Learning Toward Optimal Similarity Metric

On the Optimization and Generalization of Two-layer Transformers with Sign Gradient Descent

State Size Independent Statistical Error Bound for Discrete Diffusion Models

Direct Distributional Optimization for Provable Alignment of Diffusion Models

Metastable Dynamics of Chain-of-Thought Reasoning: Provable Benefits of Search, RL and Distillation

Degrees of Freedom for Linear Attention: Distilling Softmax Attention with Optimal Feature Efficiency

Trained Mamba Emulates Online Gradient Descent in In-Context Linear Regression

Generalization Bound of Gradient Flow through Training Trajectory and Data-dependent Kernel

How Does Label Noise Gradient Descent Improve Generalization in the Low SNR Regime?

Optimality and Adaptivity of Deep Neural Features for Instrumental Variable Regression

Nonlinear transformers can perform inference-time feature learning

From Shortcut to Induction Head: How Data Diversity Shapes Algorithm Selection in Transformers

Provable In-Context Vector Arithmetic via Retrieving Task Concepts

On the Role of Label Noise in the Feature Learning Process

When Does Metadata Conditioning (NOT) Work for Language Model Pre-Training? A Study with Context-Free Grammars

Flow matching achieves almost minimax optimal convergence

State Space Models are Provably Comparable to Transformers in Dynamic Token Selection

Quantifying Memory Utilization with Effective State-Size

Label Noise Gradient Descent Improves Generalization in the Low SNR Regime

Propagation of Chaos for Mean-Field Langevin Dynamics and its Application to Model Ensemble

Mixture of Experts Provably Detect and Learn the Latent Cluster Structure in Gradient-Based Learning

The Role of Label Noise in the Feature Learning Process

Direct Density Ratio Optimization: A Statistically Consistent Approach to Aligning Large Language Models

Quantifying Memory Utilization with Effective State-Size