Timezone: »

 
Poster
ACIL: Analytic Class-Incremental Learning with Absolute Memorization and Privacy Protection
HUIPING ZHUANG · Zhenyu Weng · Hongxin Wei · RENCHUNZI XIE · Kar-Ann Toh · Zhiping Lin

@

Class-incremental learning (CIL) learns a classification model with training data of different classes arising progressively. Existing CIL either suffers from serious accuracy loss due to catastrophic forgetting, or invades data privacy by revisiting used exemplars. Inspired by learning of linear problems, we propose an analytic class-incremental learning (ACIL) with absolute memorization of past knowledge while avoiding breaching of data privacy (i.e., without storing historical data). The absolute memorization is demonstrated in the sense that the CIL using ACIL given present data would give identical results to that from its joint-learning counterpart that consumes both present and historical samples. This equality is theoretically validated. The data privacy is ensured by showing that no historical data are involved during the learning process. Empirical validations demonstrate ACIL's competitive accuracy performance with near-identical results for various incremental task settings (e.g., 5-50 phases). This also allows ACIL to outperform the state-of-the-art methods for large-phase scenarios (e.g., 25 and 50 phases).

Keywords: Class incremental learning · recursive · exemplar-free · large-phase · absolute memorization · privacy protection
[ Paper]  [ Poster]  [ OpenReview] 

Author Information

HUIPING ZHUANG (South China University of Technology)
Zhenyu Weng (Nanyang Technological University)
Hongxin Wei (Nanyang Technological University)
RENCHUNZI XIE (Nanyang Technological University)
Kar-Ann Toh (Yonsei University)
Kar-Ann Toh

Kar-Ann Toh received the Ph.D. degree from Nanyang Technological University (NTU), Singapore. He then worked for two years in the aerospace industry, prior to his postdoctoral appointments at research centers in NTU, from 1998 to 2002. He was with the Institute for Infocomm Research, Singapore, from 2002 to 2005. He is a full Professor with the School of Electrical and Electronic Engineering, Yonsei University, South Korea. His research interests include biometrics, pattern classification, machine learning, optimization, and neural networks. Besides being active in publication, he has served as a member of advisory board and technical program committee for international conferences related to biometrics and artificial intelligence. He has served as an Associate Editor for the IEEE Transactions on Information Forensics and Security from 2013 to 2016, the Pattern Recognition Letters from 2011 to 2021, and the Journal of The Franklin Institute from 2018 to 2020. He is currently an Associate Editor of the IEEE Transactions on Biometrics, Behavior, and Identity Science, and an Editorial Board Member of IET Biometrics.

Zhiping Lin (Nanyang Technological University)

More from the Same Authors