Delta XAI

Delta-related research spans several AI subfields, focusing on improving efficiency and interpretability. Key areas include developing novel algorithms for efficient model compression (e.g., Delta-LoRA, Delta-CoMe) and creating explainable AI (XAI) methods that provide local explanations of model predictions (e.g., delta-XAI). These advancements aim to address challenges in deploying large models, enhancing their trustworthiness, and improving the understanding of their decision-making processes. The impact is seen across various applications, from video compression and robotics to legal case retrieval and continual learning.