LMUFormer: Low Complexity Yet Powerful Spiking Model With Legendre Memory Units

TL;DR

LMUFormer is a low-complexity, high-performance spiking model that combines Legendre Memory Units with convolutional modules, achieving near state-of-the-art accuracy with significantly reduced parameters and computational cost for sequence learning tasks.

Contribution

The paper introduces LMUFormer, a novel architecture that enhances recurrent models with convolutional modules and a spiking version, balancing performance and efficiency for streaming applications.

Findings

Achieves comparable performance to SOTA transformers on SCv2 dataset.

Reduces parameters by 53 times and FLOPs by 65 times compared to transformer models.

Enables real-time data processing with a 32.03% reduction in sequence length.

Abstract

Transformer models have demonstrated high accuracy in numerous applications but have high complexity and lack sequential processing capability making them ill-suited for many streaming applications at the edge where devices are heavily resource-constrained. Thus motivated, many researchers have proposed reformulating the transformer models as RNN modules which modify the self-attention computation with explicit states. However, these approaches often incur significant performance degradation. The ultimate goal is to develop a model that has the following properties: parallel training, streaming and low-cost inference, and SOTA performance. In this paper, we propose a new direction to achieve this goal. We show how architectural modifications to a recurrent model can help push its performance toward Transformer models while retaining its sequential processing capability. Specifically, inspired by the recent success of Legendre Memory Units (LMU) in sequence learning tasks, we propose LMUFormer, which augments the LMU with convolutional patch embedding and convolutional channel mixer. Moreover, we present a spiking version of this architecture, which introduces the benefit of states within the patch embedding and channel mixer modules while simultaneously reducing the computing complexity. We evaluated our architectures on multiple sequence datasets. In comparison to SOTA transformer-based models within the ANN domain on the SCv2 dataset, our LMUFormer demonstrates comparable performance while necessitating a remarkable 53 times reduction in parameters and a substantial 65 times decrement in FLOPs. Additionally, owing to our model's proficiency in real-time data processing, we can achieve a 32.03% reduction in sequence length, all while incurring an inconsequential decline in performance. Our code is publicly available at https://github.com/zeyuliu1037/LMUFormer.git.