Life-Long Deep Learning using Bayesian Principles

2022 Fiscal Year Annual Research Report

• Project/Area Number: 20H04247
• Research Institution: Institute of Physical and Chemical Research
• Principal Investigator: Khan Emtiyaz 国立研究開発法人理化学研究所, 革新知能統合研究センター, チームリーダー (30858022)
• Co-Investigator(Kenkyū-buntansha): Alquier Pierre 国立研究開発法人理化学研究所, 革新知能統合研究センター, 研究員 (10865645) [Withdrawn] ; 横田 理央 東京工業大学, 学術国際情報センター, 教授 (20760573)
• Project Period (FY): 2020-04-01 – 2023-03-31
• Keywords: continual learning / deep learning

Outline of Annual Research Achievements

Our goal was to design AI systems that continue to learn and improve throughout their lifetime. This fiscal year we worked on concluding our work and scale up our algorithm on ImageNet data. We wrote one research paper on this topic which is currently under submission.

- (Under submission) Improving Continual Learning by Accurate Gradient Reconstructions of the Past, Erik Daxberger, Siddharth Swaroop, Kazuki Osawa, Rio Yokota, Richard E Turner, Jose; Miguel Hernandez-Lobato, Mohammad Emtiyaz Khan

This work uses our previous work to propose a new improvement in continual learning. It essentially combines two methods to get state-of-the-art performance at the tiny ImageNet level.

Research Progress Status

令和4年度が最終年度であるため、記入しない。

Strategy for Future Research Activity

令和4年度が最終年度であるため、記入しない。