2 papers
cs.LG2024
Harnessing Neural Unit Dynamics for Effective and Scalable Class-Incremental Learning
Depeng Li, Tianqi Wang, Junwei Chen +2
Class-incremental learning (CIL) aims to train a model to learn new classes from non-stationary data streams without forgetting old ones. In this paper, we propose a new kind of co…
cs.LG2024
Towards Continual Learning Desiderata via HSIC-Bottleneck Orthogonalization and Equiangular Embedding
Depeng Li, Tianqi Wang, Junwei Chen +3
Deep neural networks are susceptible to catastrophic forgetting when trained on sequential tasks. Various continual learning (CL) methods often rely on exemplar buffers or/and netw…