AI & Fundamentals

Efficient and Long-Context GenAI Models - Sung Ju Hwang, KAIST Endowed Chair Professor, Kim Jaechul School of Artificial Intelligence and School of Computing at KAIST

Zoom Link

Abstract:

Generative AI models have demonstrated remarkable performance across various tasks in recent years. However, their deployment in real-world production-grade AI systems is hindered by significant computational costs. Additionally, these models are constrained by their limited context window size, which restricts their ability to handle more complex tasks that require larger working memory. In this talk, I will present recent advancements from our group aimed at addressing these challenges in generative AI models. Specifically, I will discuss methods to improve training and inference efficiency, as well as techniques for expanding the context window size.

Bio:

Dr. Sung Ju Hwang is a KAIST Endowed Chair Professor in the Kim Jaechul School of Artificial Intelligence and School of Computing at KAIST. Prior to working at KAIST, he was an Assistant Professor in the School of Electric and Computer Engineering at UNIST, and before that a Postdoctoral Research Associate at Disney Research, working under the supervision of Professor Leonid Sigal. Prof. Hwang did his Ph.D. in computer science at University of Texas at Austin, under the supervision of Professor Kristen Grauman. During his Ph.D. he also closely collaborated with Professor Fei Sha at University of Southern California. Prof. Hwang is also the CEO of DeepAuto, an Enterprise AI startup building an Automated, Low-cost, and Fast Generative AI Platform for Enterprises.
 

Zoom Link