Quality Alchemist (品質煉金術師): COMP Distinguished Seminar – (Re)visiting Foundation Models for Science and Beyond

2024年12月24日星期二

COMP Distinguished Seminar – (Re)visiting Foundation Models for Science and Beyond

Prof. Wei Wang, who is the Leonard Kleinrock Chair Professor in Computer Science and Computational Medicine at University of California, Los Angeles and the director of the Scalable Analytics Institute (ScAi), was invited as our guest speaker for COMP Distinguished Seminar Series on Data Science & Artificial Intelligence. Her topic entitled “(Re) visiting Foundation Models for Science and Beyond”.

In the beginning, Prof. Wei Wang introduced some foundation models including GPT-4, CLIP and SAM, etc.

Then she briefed Large Language Models’ evolutionary tree and (Pre-)trained Foundation Models with three major technique types.

And then she compared different LLM’s problem solving skills and found inaccuracy in different cases. For instance, clinical decision capabilities using GPT-40 is better than open models.

Some error analyses were demonstrated among Llama, GPT-3.5 and GPT-4o. Error analysis of zero-shot learning (ZSL), ZSL + CoT prompting, Few-Shot Learning + Python, …, were also mentioned.

When discussed Medical LLM, bias analysis was concerned.

After that multi-modal LLMs for clinical application were discussed.

Finally, LLMs integrated with Visual Knowledge Graph was demonstrated in diagnosis prediction.

Lastly, Prof. Prof. Wei Wang discussed different application of LLM such as MM-AIM for discovery and advanced manufacturing of extreme materials. I remembered using a fuzzy logic expert system for material selection on my final year project (FYP) long time ago.