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.
arXiv commentAccepted to NeurIPS 2021
References in corpus (30)
- Adam: A Method for Stochastic Optimization
- PyTorch: An Imperative Style, High-Performance Deep Learning Library
- Playing Atari with Deep Reinforcement Learning
- Dota 2 with Large Scale Deep Reinforcement Learning
- Decision Transformer: Reinforcement Learning via Sequence Modeling
- Massively Parallel Methods for Deep Reinforcement Learning
- Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention
- Learning to Execute
- Agent57: Outperforming the Atari Human Benchmark
- Language Modeling with Deep Transformers
- Using Fast Weights to Attend to the Recent Past
- Theoretical Limitations of Self-Attention in Neural Sequence Models
- Random Feature Attention
- Pretrained Transformers as Universal Computation Engines
- Psychlab: A Psychology Laboratory for Deep Reinforcement Learning Agents
- LambdaNetworks: Modeling Long-Range Interactions Without Attention
- Enhancing the Transformer with Explicit Relational Encoding for Math Problem Solving
- Backpropamine: training self-modifying neural networks with differentiable neuromodulated plasticity
- Addressing Some Limitations of Transformers with Feedback Memory
- TorchBeast: A PyTorch Platform for Distributed RL
- Linear Transformers Are Secretly Fast Weight Programmers
- Training Agents using Upside-Down Reinforcement Learning
- Reinforcement Learning Upside Down: Don't Predict Rewards -- Just Map Them to Actions
- Dropout is a special case of the stochastic delta rule: faster and more accurate deep learning
- Learning advanced mathematical computations from examples
- Efficient Transformers in Reinforcement Learning using Actor-Learner Distillation
- The Devil is in the Detail: Simple Tricks Improve Systematic Generalization of Transformers
- The Neural Data Router: Adaptive Control Flow in Transformers Improves Systematic Generalization
- Training and Generating Neural Networks in Compressed Weight Space
- Fast Weight Long Short-Term Memory