March 14, 2024, 4:42 a.m. | Shengzhuang Chen, Jihoon Tack, Yunqiao Yang, Yee Whye Teh, Jonathan Richard Schwarz, Ying Wei

cs.LG updates on

arXiv:2403.08477v1 Announce Type: cross
Abstract: Conventional wisdom suggests parameter-efficient fine-tuning of foundation models as the state-of-the-art method for transfer learning in vision, replacing the rich literature of alternatives such as meta-learning. In trying to harness the best of both worlds, meta-tuning introduces a subsequent optimization stage of foundation models but has so far only shown limited success and crucially tends to underperform on out-of-domain (OOD) tasks. In this paper, we introduce Sparse MetA-Tuning (SMAT), a method inspired by sparse mixture-of-experts …

abstract art arxiv best of cs.lg experts few-shot fine-tuning foundation harness literature meta meta-learning optimization power stage state through transfer transfer learning type vision

Senior Data Engineer

@ Displate | Warsaw

Professor/Associate Professor of Health Informatics [LKCMedicine]

@ Nanyang Technological University | NTU Novena Campus, Singapore

Research Fellow (Computer Science (and Engineering)/Electronic Engineering/Applied Mathematics/Perception Sciences)

@ Nanyang Technological University | NTU Main Campus, Singapore

Java Developer - Assistant Manager

@ State Street | Bengaluru, India

Senior Java/Python Developer

@ General Motors | Austin IT Innovation Center North - Austin IT Innovation Center North

Research Associate (Computer Engineering/Computer Science/Electronics Engineering)

@ Nanyang Technological University | NTU Main Campus, Singapore