🏢 Georgia Tech

R³M enhances reinforcement learning from human feedback by robustly handling corrupted preference labels, consistently learning the underlying reward and identifying outliers with minimal computationa…

RefDrop: A training-free method enhances image and video generation consistency by directly controlling the influence of reference features on the diffusion process, enabling precise manipulation of c…

RankRAG: One LLM, dual-purpose instruction-tuning for superior RAG!

Researchers discover ‘safety basins’ in LLMs, proposing a new metric (VISAGE) to quantify finetuning risks and visualize how these basins protect against safety compromise during model training.

Diffusion Spectral Representation (Diff-SR) enables efficient reinforcement learning by extracting sufficient value function representations from diffusion models, bypassing slow sampling and facilita…