Timezone: »

Exploring Example Influence in Continual Learning
Qing Sun · Fan Lyu · Fanhua Shang · Wei Feng · Liang Wan


Continual Learning (CL) sequentially learns new tasks like human beings, with the goal to achieve better Stability (S, remembering past tasks) and Plasticity (P, adapting to new tasks). Due to the fact that past training data is not available, it is valuable to explore the influence difference on S and P among training examples, which may improve the learning pattern towards better SP. Inspired by Influence Function (IF), we first study example influence via adding perturbation to example weight and computing the influence derivation. To avoid the storage and calculation burden of Hessian inverse in neural networks, we propose a simple yet effective MetaSP algorithm to simulate the two key steps in the computation of IF and obtain the S- and P-aware example influence. Moreover, we propose to fuse two kinds of example influence by solving a dual-objective optimization problem, and obtain a fused influence towards SP Pareto optimality. The fused influence can be used to control the update of model and optimize the storage of rehearsal. Empirical results show that our algorithm significantly outperforms state-of-the-art methods on both task- and class-incremental benchmark CL datasets.

Author Information

Qing Sun (Tianjin University)
Fan Lyu (Tianjin University)
Fanhua Shang (Tianjin University)
Wei Feng (Tianjin University)

Wei Feng received Ph.D. degree in computer science from City University of Hong Kong in 2008. He is currently a full professor in the School of Computer Science and Technology, Tianjin University. His major research interest is active robotic vision and visual intelligence, specifically including active camera relocalization and lighting recurrence, general Markov Random Fields modeling, dis- crete/continuous energy minimization, image segmentation, active 3D scene perception, SLAM, and generic pattern recognition. Recently, he focuses on solving preventive conservation problems of cultural heritages via computer vision and machine learning. He is an associate editor of Neurocomputing and JAIHC.

Liang Wan (Tianjin University)

More from the Same Authors