Focus Your Attention (with Adaptive IIR Filters)
This work addresses efficiency and performance challenges in long-range sequence modeling for machine learning applications, representing an incremental improvement over existing methods.
The authors tackled the problem of improving attention mechanisms in neural networks by introducing a new layer that uses dynamic Infinite Impulse Response (IIR) filters to preprocess input sequences, achieving performance on-par with state-of-the-art models like Heyna, GPT2, and Mega while using fewer parameters and sub-quadratic time complexity.
We present a new layer in which dynamic (i.e.,input-dependent) Infinite Impulse Response (IIR) filters of order two are used to process the input sequence prior to applying conventional attention. The input is split into chunks, and the coefficients of these filters are determined based on previous chunks to maintain causality. Despite their relatively low order, the causal adaptive filters are shown to focus attention on the relevant sequence elements. The new layer is grounded in control theory, and is shown to generalize diagonal state-space layers. The layer performs on-par with state-of-the-art networks, with a fraction of their parameters and with time complexity that is sub-quadratic with input size. The obtained layer is favorable to layers such as Heyna, GPT2, and Mega, both with respect to the number of parameters and the obtained level of performance on multiple long-range sequence problems.