Mathematical Approaches to Interpreting and Optimizing Large Language Models
|
Organizer(s): |
Name:
|
Affiliation:
|
Country:
|
Dejing Dou
|
BCG X
|
Peoples Rep of China
|
Xianfeng Gu
|
Stony Brook University
|
USA
|
|
|
|
|
|
|
|
|
Introduction:
| | Most Large Language Models (LLMs), although are very successful so far, still have a range of issues that need to be addressed both by academic and industry, such as deepfakes, misinformation, energy efficiency, data privacy, bias and fairness etc. Addressing these issues requires a combination of research and technology breakthroughs especially in improving interpretability and efficiency of LLMs during pre-training, fine-tuning and prompt engineering stages. Among various approaches to address those issues, mathematical and statistical approaches are very promising and worth deep investigations. By highlighting those important issues and approaches, the session aims to stimulate dialogue, collaboration, and innovation, ultimately driving forward both methodological development and practical applications in interpreting and optimizing LLMs. |
|
|
|
|
|
