CSE - Making Better Foundation Models

Foundation Models (FMs), such as large language models, graph generative models, multimodal models, vision-language-action models, and world models, are being used in many places. Our lab is looking for self-motivated undergraduate students that are interested in developing better FMs or finding better methods to use FMs. We study FMs in various real-world scenarios from personalizing intelligent assistance, advancing scientific discovery, designing circuits, interacting with physical environments, to improving privacy, safety, and alignment with virtue ethics. We aim to improve several types of abilities of FMs: (a) follow complex instructions, (b) perform complex reasoning and tool use, (c) detect and reduce undesired behaviors, (d) unlearn to forget private or copyright data, (e) understand and generate multimodal data, (f) provide personalized responses, (g) align model behaviors with ethical standards, etc.