Practical Computational Power of Linear Transformers and Their Recurrent and Self-Referential Extensions
Submission Type: Regular Short Paper
Submission Track: Machine Learning for NLP
Keywords: recurrent neural networks, RNNs, transformers, computational power, automata, counter machines, formal languages, linear transformers, self-reference, self-referential weight matrix
TL;DR: We study the practical computational power of linear Transformers and their extensions using formal languages
Abstract: Recent studies of the computational power of recurrent neural networks (RNNs) reveal a hierarchy of RNN architectures, given real-time and finite-precision assumptions. Here we study auto-regressive Transformers with linearised attention, a.k.a. linear Transformers (LTs) or Fast Weight Programmers (FWPs). LTs are special in the sense that they are equivalent to RNN-like sequence processors with a fixed-size state, while they can also be expressed as the now-popular self-attention networks. We show that many well-known results for the standard Transformer directly transfer to LTs/FWPs. Our formal language recognition experiments demonstrate how recently proposed FWP extensions such as recurrent FWPs and self-referential weight matrices successfully overcome certain limitations of the LT, e.g., allowing for generalisation on the parity problem. Our code is public.
Submission Number: 3713
Loading