Silvio Savarese

Dr. Silvio Savarese is an Adjunct Professor in the Department of Computer Science at Stanford University. His research focuses on advancing artificial intelligence, particularly in multimodal models, autonomous agents, and large language models (LLMs). He has contributed to foundational work in vision-language models like BLIP3 and XGen, as well as frameworks for training action models and optimizing AI agents. His projects emphasize practical applications such as CRM systems, code generation, and ethical AI. Current interests include scalable reasoning in LLMs, efficient model architectures, and multimodal pretraining for 3D understanding.

Key research directions include developing unified multimodal models capable of handling vision, language, and point cloud data, as seen in projects like ULIP and Merlion. He explores agent systems through benchmarks like Behavior-1K and BOLAA, aiming to improve real-world interaction capabilities. Recent work addresses challenges in data generation (Text2Data), agent optimization (PRACT), and safety through methods like LZ Penalty for text repetition control.

His contributions span open-source libraries (LAVIS, OMNIXAI) and datasets (DialogStudio, Eye-BEHAVIOR), emphasizing reproducibility and community impact. While no formal awards are listed, his extensive publication record and industry collaborations highlight academic and industrial relevance. Current projects include high-fidelity video synthesis (xGen-videosyn-1), time series forecasting (Moirai-MoE), and lightweight agent frameworks (AgentLite).