[행사/세미나] 인공지능대학원 전문가 초청 세미나 (KAIST 박사후연구원 강해용박사)
- 소프트웨어학과
- 조회수2603
- 2024-07-11
제목: Tutorials & Subnetworks for Continual Learning
연사: 강해용박사
시간: 2024년 7월 17일(수) 14시
방식: 오프라인 (제2공학관 26동 26312호)
요약: In the first part, to understand better Continual Learning (CL), this work provides a tutorial on CL, including four kinds of CL methods: (1) regularized-based, (2) replay-based, (3) architecture-based, and (4) prompt-based methods. In the second part, Inspired by the Lottery Ticket Hypothesis (LTH), which highlights the existence of efficient subnetworks within larger, dense networks, high-performing Winning Subnetworks (WSN [ICML2022], Soft-SubNetwork (SoftNet)[ICLR2023], and Progressive Fourier Neural Representation (PFNR) [ICLR2024]) are considered for various continual learning scenarios. Lastly, we briefly discuss the current continual learning research trend.
이력: Haeyong Kang is currently a postdoctoral researcher at KAIST Electrical Engineering. He completed his Ph.D. with a dissertation on "Forget-free Subnetworks for Life-long Learning" in August 2023. His research interests include Sparse/Pruned Neural Networks (ResNets and Transformers), Continual/Incremental Learning, Multimodal Representation, and Imbalanced Dataset Learning.