Deepak Pathak

Latent Particle World Models: Self-supervised Object-centric Stochastic Dynamics Modeling

ViPRA: Video Prediction for Robot Actions

From Tokens to Actions: Discrete Flow Matching for Visuomotor Policy Learning

Self-Questioning Language Models

Maximizing Confidence Alone Improves Reasoning

Evolutionary Policy Optimization

Diffusion Beats Autoregressive in Data-Constrained Settings

Generative Classifiers Avoid Shortcut Solutions

VADER: Video Diffusion Alignment via Reward Gradients

Unified Multimodal Discrete Diffusion