Timezone: »
One of the objectives of continual learning is to prevent catastrophic forgetting in learning multiple tasks sequentially,and the existing solutions have been driven by the conceptualization of the plasticity-stability dilemma.However, the convergence of continual learning for each sequential task is less studied so far.In this paper, we provide a convergence analysis of memory-based continual learning with stochastic gradient descentand empirical evidence that training current tasks causes the cumulative degradation of previous tasks.We propose an adaptive method for nonconvex continual learning (NCCL), which adjusts step sizes of both previous and current tasks with the gradients.The proposed method can achieve the same convergence rate as the SGD method when the catastrophic forgetting term which we define in the paper is suppressed at each iteration.Further, we demonstrate that the proposed algorithm improves the performance of continual learning over existing methods for several image classification tasks.
Author Information
Seungyub Han (Seoul National University)
Yeongmo Kim (Seoul National University)
Taehyun Cho (seoul university)
Jungwoo Lee (Seoul National University)
More from the Same Authors
-
2022 : Perturbed Quantile Regression for Distributional Reinforcement Learning »
Taehyun Cho · Seungyub Han · Heesoo Lee · Kyungjae Lee · Jungwoo Lee -
2023 Poster: Pitfall of Optimism: Distributional Reinforcement Learning by Randomizing Risk Criterion »
Taehyun Cho · Seungyub Han · Heesoo Lee · Kyungjae Lee · Jungwoo Lee -
2023 Poster: SPQR: Controlling Q-ensemble Independence for Reinforcement Learning »
Dohyeok Lee · Seungyub Han · Taehyun Cho · Jungwoo Lee -
2022 : Poster Session 1 »
Andrew Lowy · Thomas Bonnier · Yiling Xie · Guy Kornowski · Simon Schug · Seungyub Han · Nicolas Loizou · xinwei zhang · Laurent Condat · Tabea E. Röber · Si Yi Meng · Marco Mondelli · Runlong Zhou · Eshaan Nichani · Adrian Goldwaser · Rudrajit Das · Kayhan Behdin · Atish Agarwala · Mukul Gagrani · Gary Cheng · Tian Li · Haoran Sun · Hossein Taheri · Allen Liu · Siqi Zhang · Dmitrii Avdiukhin · Bradley Brown · Miaolan Xie · Junhyung Lyle Kim · Sharan Vaswani · Xinmeng Huang · Ganesh Ramachandra Kini · Angela Yuan · Weiqiang Zheng · Jiajin Li -
2018 : Poster session »
David Zeng · Marzieh S. Tahaei · Shuai Chen · Felix Meister · Meet Shah · Anant Gupta · Ajil Jalal · Eirini Arvaniti · David Zimmerer · Konstantinos Kamnitsas · Pedro Ballester · Nathaniel Braman · Udaya Kumar · Sil C. van de Leemput · Junaid Qadir · Hoel Kervadec · Mohamed Akrout · Adrian Tousignant · Matthew Ng · Raghav Mehta · Miguel Monteiro · Sumana Basu · Jonas Adler · Adrian Dalca · Jizong Peng · Sungyeob Han · Xiaoxiao Li · Karthik Gopinath · Joseph Cheng · Bogdan Georgescu · Kha Gia Quach · Karthik Sarma · David Van Veen