일시: 2025. 6. 9 (월요일), 오후 2시~3시

장소: 제1공학관, A528

Speaker: Youngeun Kim Ph.D. (Research Scientist at Meta Reality Labs)

Title: Towards Efficient Deep Learning: Brain-Inspired Algorithm, Fine-Tuning, and Compression

Abstract:

As neural networks grow in size and complexity, improving their efficiency has become critical for practical deployment and broader accessibility. In this talk, I will present a holistic view of techniques for enhancing the efficiency of deep neural networks across training and inference stages. I will begin with brain-inspired machine learning algorithms that open up new directions for energy-efficient and robust model design, taking cues from biological systems. Then, I will introduce a lightweight and effective fine-tuning strategy that enables adaptation without updating the full weight parameters inside a model. Finally, I will discuss model quantization techniques that significantly reduce computational and memory overhead for deployment, particularly on edge devices. Together, these approaches contribute to building scalable, efficient, and intelligent AI systems.

Bio:

Youngeun Kim is a Research Scientist at Meta Reality Labs, working on neural network systems for EMG-based human-computer interaction. He received his Ph.D. in Electrical & Computer Engineering from Yale University in 2024, advised by Prof. Priyadarshini Panda. He also holds an M.S. in Electrical Engineering from KAIST (2020) and a B.S. in Electronic Engineering from Sogang University (2018). His industry experience includes AI research internships at Kakao Corporation, SK Telecom T-brain, and Samsung Advanced Institute of Technology (SAIT) in South Korea, as well as an Applied Scientist internship at Amazon AWS AI in Seattle. His research interests span efficient machine learning algorithms, neuromorphic computing, computer vision, and algorithm-hardware co-design.