Editorial: Understanding and bridging the gap between neuromorphic computing and machine learning, volume II

Lei Deng, Huajin Tang, Kaushik RoyFrontiers Comput. Neurosci 2024 · OpenReview

Introduction

The quest for artificial intelligence has long been a central goal of human innovation, leading to the development of two distinct yet increasingly intersecting paths: neuromorphic computing and machine learning. Neuromorphic computing, driven by neuroscience, focuses on spiking neural networks (SNNs) and specialized hardware to mimic the brain's energy efficiency. In contrast, machine learning, driven by computer science, utilizes artificial neural networks (ANNs) and powerful accelerators like GPUs to achieve high accuracy in complex tasks.

Despite its efficiency, neuromorphic computing remains in its early stages, facing a significant "gap" in accuracy and application compared to the mature machine learning ecosystem. This editorial summarizes the second volume of a Research Topic dedicated to bridging this divide, featuring 11 papers that explore neuromorphic models, hardware implementations, and programming frameworks.

Advancing Neuromorphic Models and Algorithms

SNNs encode information through discrete spike events and neural dynamics, distinguishing them from the continuous activations of ANNs. However, their non-differentiable nature has historically made them difficult to train using standard gradient-based methods. To overcome this, researchers are now integrating sophisticated backpropagation techniques from machine learning into the SNN domain.

One prominent approach discussed in this volume is the use of three-factor Hebbian updates. By employing auxiliary memory, these algorithms can bridge the gap between local synaptic changes and global error signals, allowing for more effective supervised learning. Furthermore, researchers have introduced adjustable synaptic delays to enhance temporal processing and SNN-specific self-attention mechanisms (STCA-SNN). These advancements enable networks to focus on relevant spatiotemporal data more effectively, improving performance on benchmarks such as spoken word recognition and image reconstruction.

Optimizing Hardware and Mapping Strategies

The sparse and event-driven nature of spikes makes standard GPUs inefficient for executing SNNs. Neuromorphic hardware is specifically designed to handle these dynamics through decentralized many-core architectures. However, these systems often suffer from fragmented memory and decentralized execution, which can lower resource utilization.

Wang et al. (2023) address this by proposing the "mapping limit" concept, which identifies the maximum potential for resource saving during the logical and physical mapping of neural networks onto neuromorphic chips. They introduced a closed-loop mapping strategy featuring an asynchronous 4D model partition and a Hamilton loop algorithm (HLA). Validated on the TianjicX neuromorphic chip, these methods demonstrate significant gains in processing efficiency, providing a foundation for more general and efficient hardware mapping frameworks.

The Role of Software Frameworks

Software is a critical component of the neuromorphic ecosystem, often determining the practical efficiency gained from hardware. The Research Topic highlights the Brain-Inspired Deep Learning (BIDL) framework, designed to bridge the gap between high-level model design and low-level hardware execution. BIDL supports a wide range of data types, including Dynamic Vision Sensor (DVS) signals, videos, and 3D medical images. By providing a user-friendly interface that optimizes performance for both GPUs and neuromorphic chips, frameworks like BIDL are essential for making neuromorphic technology accessible to the broader research community.

Conclusion and Future Directions

While neuromorphic systems have yet to demonstrate superior performance over traditional machine learning in mainstream intelligent tasks, the research presented here marks a significant step forward. The development of an iterative ecosystem—where algorithms, hardware, and software inform and improve one another—is key to closing the performance gap. As we continue to refine these hybrid systems, the ultimate goal of achieving brain-like intelligence becomes increasingly attainable. This Research Topic serves as a vital contribution to the ongoing dialogue between neuroscience and computer science, paving the way for the next generation of intelligent systems.

Summary updated

Abstract

Pursuing intelligence is a long-term goal of the human, towards which two routes have been paved on the road: neuromorphic computing driven by neuroscience and machine learning driven by computer science (Pei et al., 2019). Spiking neural networks (SNNs) and neuromorphic chips (Basu et al., 2022;Christensen et al., 2022) dominate the neuromorphic computing domain, while artificial neural networks (ANNs) and machine learning accelerators (Deng et al., 2020) dominate the machine learning domain. Neuromorphic computing with efficient models and hardware has shown energy efficiency superiority (Renner et al., 2021), however, still lies in its infant stage and presents a gap in terms of accuracy and applications compared to the mature machine learning ecosystem.To this end, we proposed a Research Topic, named "Understanding and Bridging the Gap between Neuromorphic Computing and Machine Learning", in Frontiers in Neuroscience and Frontiers in Computational Neuroscience in 2019, and have successfully published 14 papers on neuromorphic computing and machine learning (Deng et al., 2021). Encouraged by such positive impetus for the neuromorphic computing community, we relaunched the Research Topic in 2022.This time, we have accepted 11 submissions in the end. The scope of these works covers neuromorphic models and algorithms, hardware implementation, and programming frameworks.SNNs encode information in spike events and process information using neural dynamics, which differ from ANNs. Due to the complicated spatiotemporal dynamics and non-differentiable spike activities, the SNN domain uses plasticitybased unsupervised learning algorithms (Diehl and Cook, 2015) for a long time but suffers from low accuracy. To break the bottleneck of lacking effective learning algorithms, the sophisticated backpropagation method in machine learning has Neuromorphic models enjoy low computational costs owing to the binary spike representation and sparse operations. However, directly executing SNNs on GPUs without tailored optimization is inefficient. Neuromorphic hardware is designed for the efficient execution of SNNs via event-driven computing (Merolla et al., 2014). The decentralized manycore architecture with high computing parallelism and memory locality is widely adopted by neuromorphic chips. However, its fragmented memories and decentralized execution lowers the resource utilization and processing efficiency. Wang et al., 2023 propose the mapping limit concept which points out the resource saving upper limit during logical and physical mapping when deploying neural networks onto neuromorphic chips. A closed-loop mapping strategy with an asynchronous 4D model partition for logical mapping and a Hamilton loop algorithm (HLA) for physical mapping are elaborated. Their methods and performance gains are validated on the TianjicX neuromorphic chip (Ma et al., 2022), which is helpful for building a general and efficient mapping framework for neuromorphic hardware.Software is one of the key components in the ecosystem of neuromorphic computing (Fang et al., 2023), which is sometimes more important than the hardware itself because it determines how much practical efficiency we can gain from the peak efficiency of hardware. In this Research Topic, we accepted one paper on the programming framework for neuromorphic models in this Research Neuromorphic computing is a neuroscience-driven domain in pursuing brain-like intelligence, which is an important route distinct from machine learning. Although neuromorphic systems have not yet demonstrated superior performance over machine learning systems in main stream intelligent tasks, we believe it can be significantly improved when the neuromorphic ecosystem is constructed and becomes iterative between algorithms, models, hardware, software, and benchmarks. This Research Topic is a quite minor step. We hope future works can really bridge the gap between neuromorphic computing and machine learning, along the way to reach the long-term goal of mimicking brain intelligence.
BibTeX
@inproceedings{deng2024editorial,
  author = {Lei Deng and Huajin Tang and Kaushik Roy},
  title = {{Editorial: Understanding and bridging the gap between neuromorphic computing and machine learning, volume II}},
  booktitle = {Frontiers Comput. Neurosci},
  year = {2024},
  url = {https://openreview.net/forum?id=85Haxt4A7e},
}
Figure