Three scenarios for continual learning
Gido M. van de Ven1,2、Andreas S. Tolias1,3 1 Center for Neuroscience and Artificial Intelligence, Baylor College of Medicine, Houston 2 Computational and Biological Learning Lab, University of Cambridge, Cambridge 3 Department of Electrical and Computer Engineering, Rice University, Houston
Abstract
Standard artificial neural networks suffer from the well-known issue of catastrophic forgetting, making continual or lifelong learning difficult for machine learning. In recent years, numerous methods have been proposed for continual learning, but due to differences in evaluation protocols it is difficult to directly compare their performance. To enable more structured comparisons, we describe three continual learning scenarios based on whether at test time task identity is provided and—in case it is not—whether it must be inferred. Any sequence of well-defined tasks can be performed according to each scenario. Using the split and permuted MNIST task protocols, for each scenario we carry out an extensive comparison of recently proposed continual learning methods. We demonstrate substantial differences between the three scenarios in terms of difficulty and in terms of how efficient different methods are. In particular, when task identity must be inferred (i.e., class incremental learning), we find that regularization-based approaches (e.g., elastic weight consolidation) fail and that replaying representations of previous experiences seems required for solving this scenario.
中文速览
深度神经网络在顺序学习多个任务时会严重"遗忘"旧知识，但学界对这一灾难性遗忘（catastrophic forgetting）问题的评测标准五花八门，导致各方法的性能根本无从直接比较。为此，这篇论文提出了三种递进难度的持续学习（continual learning）场景——任务增量学习（Task-IL，测试时告知任务编号）、域增量学习（Domain-IL，不告知编号但无需推断）、类增量学习（Class-IL，既不告知也必须自行推断），并在分割与置换 MNIST 基准上对弹性权重巩固（EWC）、突触智能（SI）、深度生成回放（DGR）、iCaRL 等主流方法逐一对比。结果表明，三个场景的难度差异显著：在最难的类增量场景下，正则化类方法彻底失效，只有依赖"回放"历史经验的方法才能维持可接受的性能。这一发现为持续学习研究提供了统一的评测框架，揭示了不同方法的真实适用边界，也指明了未来攻关的核心方向。
原文 arXiv:1904.07734;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1904.07734v1