Akshay Krishnamurthy

The Coverage Principle: How Pre-Training Enables Post-Training

A Unifying View of Coverage in Linear Off-policy Evaluation

Representation-Based Exploration for Language Models: From Test-Time to Post-Training

Let's Let's Let's Let's... Understand Looping in Reasoning Models

Computationally Efficient RL under Linear Bellman Completeness for Deterministic Dynamics

Self-Improvement in Language Models: The Sharpening Mechanism

Is Best-of-N the Best of Them? Coverage, Scaling, and Optimality in Inference-Time Alignment

Exploratory Preference Optimization: Harnessing Implicit Q*-Approximation for Sample-Efficient RLHF

Correcting the Mythos of KL-Regularization: Direct Alignment without Overoptimization via Chi-Squared Preference Optimization