Going Beyond Linear Transformers with Recurrent Fast Weight Programmers
This work addresses the need for more flexible and powerful sequence models in AI, particularly for tasks like reinforcement learning and language modeling, though it is incremental as it builds on existing Fast Weight Programmer frameworks.
The authors tackled the problem of extending linear Transformers by adding recurrence to both slow and fast networks in Fast Weight Programmers, resulting in models that combine Transformer and RNN properties, with large improvements over LSTM in several Atari games.
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.