Continual Learning in Recurrent Neural Networks
arXiv:2006.12109
Abstract
While a diverse collection of continual learning (CL) methods has been proposed to prevent catastrophic forgetting, a thorough investigation of their effectiveness for processing sequential data with recurrent neural networks (RNNs) is lacking. Here, we provide the first comprehensive evaluation of established CL methods on a variety of sequential data benchmarks. Specifically, we shed light on the particularities that arise when applying weight-importance methods, such as elastic weight consolidation, to RNNs. In contrast to feedforward networks, RNNs iteratively reuse a shared set of weights and require working memory to process input samples. We show that the performance of weight-importance methods is not directly affected by the length of the processed sequences, but rather by high working memory requirements, which lead to an increased need for stability at the cost of decreased plasticity for learning subsequent tasks. We additionally provide theoretical arguments supporting this interpretation by studying linear RNNs. Our study shows that established CL methods can be successfully ported to the recurrent case, and that a recent regularization approach based on hypernetworks outperforms weight-importance methods, thus emerging as a promising candidate for CL in RNNs. Overall, we provide insights on the differences between CL in feedforward networks and RNNs, while guiding towards effective solutions to tackle CL on sequential data.
arXiv commentPublished at ICLR 2021
References in corpus (26)
- Auto-Encoding Variational Bayes
- Distilling the Knowledge in a Neural Network
- Overcoming catastrophic forgetting in neural networks
- WaveNet: A Generative Model for Raw Audio
- Spectral Normalization for Generative Adversarial Networks
- Pixel Recurrent Neural Networks
- Continual Learning Through Synaptic Intelligence
- Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shift
- A Recurrent Latent Variable Model for Sequential Data
- Three scenarios for continual learning
- FearNet: Brain-Inspired Model for Incremental Learning
- Alleviating catastrophic forgetting using context-dependent gating and synaptic stabilization
- Learning Stochastic Recurrent Networks
- Measuring Catastrophic Forgetting in Neural Networks
- Generative replay with feedback connections as a general strategy for continual learning
- On orthogonality and learning recurrent networks with long term dependencies
- On Quadratic Penalties in Elastic Weight Consolidation
- Task Agnostic Continual Learning via Meta Learning
- A Unifying Bayesian View of Continual Learning
- Memory Augmented Neural Networks with Wormhole Connections
- Continual Learning with Gated Incremental Memories for sequential data processing
- Toward Continual Learning for Conversational Agents
- Progressive Memory Banks for Incremental Domain Adaptation
- Incremental Sequence Learning
- Evaluating Online Continual Learning with CALM
- Making Good on LSTMs' Unfulfilled Promise