Going Beyond Linear Transformers with Recurrent Fast Weight Programmers
arXiv:2106.06295
Abstract
Transformers with linearised attention (''linear Transformers'') have demonstrated the practical scalability and effectiveness of outer product-based Fast Weight Programmers (FWPs) from the '90s. However, the original FWP formulation is more general than the one of linear Transformers: a slow neural network (NN) continually reprograms the weights of a fast NN with arbitrary architecture. In existing linear Transformers, both NNs are feedforward and consist of a single layer. Here we explore new variations by adding recurrence to the slow and fast nets. We evaluate our novel recurrent FWPs (RFWPs) on two synthetic algorithmic tasks (code execution and sequential ListOps), Wikitext-103 language models, and on the Atari 2600 2D game environment. Our models exhibit properties of Transformers and RNNs. In the reinforcement learning setting, we report large improvements over LSTM in several Atari games. Our code is public.
Accepted to NeurIPS 2021
References in corpus (9)
- PyTorch: An Imperative Style, High-Performance Deep Learning Library
- Decision Transformer: Reinforcement Learning via Sequence Modeling
- Massively Parallel Methods for Deep Reinforcement Learning
- Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention
- Using Fast Weights to Attend to the Recent Past
- Random Feature Attention
- LambdaNetworks: Modeling Long-Range Interactions Without Attention
- TorchBeast: A PyTorch Platform for Distributed RL
- Efficient Transformers in Reinforcement Learning using Actor-Learner Distillation