Basura Fernando

H-GRPO: Permutation-Invariant Reinforcement Learning for Grounded Visual Reasoning

Eric Peh, Debaditya Roy, Basura Fernando

Vision-Language Models (VLMs) often achieve high performance on benchmarks while remaining "black boxes", yet they remain prone to hallucination or rely on superficial shortcuts. In this work, we propose a framework designed to enhance both performance and interpretability throug…Read more