Dileep Kalathil

Diffusion Blend: Inference-Time Multi-Preference Alignment for Diffusion Models

Curriculum Reinforcement Learning from Easy to Hard Tasks Improves LLM Reasoning

Inference-Time Search using Side Information for Diffusion-based Image Reconstruction

MAVIS: Multi-Objective Alignment via Value-Guided Inference-Time Search

Robust LLM Alignment via Distributionally Robust Direct Preference Optimization