Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Asynchronous Bioplausible Neuron for SNN for Event Vision

Sanket Kachole, Hussain Sajwani, Fariborz Baghaei Naeini, Dimitrios Makris, Yahya Zweiri

TL;DR

The paper tackles the challenge of maintaining homeostasis and energy efficiency in spiking neural networks for event-based vision. It introduces the Asynchronous Bioplausible Neuron (ABN), a per-neuron dynamic-threshold mechanism combining Membrane Gradient, Threshold Retrospective Gradient, and Spike Efficiency, integrated into a spiking MLP with STDP. Across six diverse datasets, ABN delivers state-of-the-art classification and segmentation accuracy, robust performance under degraded inputs, strong homeostasis, and lower energy consumption compared to prior methods. The findings highlight ABN's potential for real-time, neuromorphic vision systems, while acknowledging hardware deployment considerations and opportunities for further biophysiological realism.

Abstract

Spiking Neural Networks (SNNs) offer a biologically inspired approach to computer vision that can lead to more efficient processing of visual data with reduced energy consumption. However, maintaining homeostasis within these networks is challenging, as it requires continuous adjustment of neural responses to preserve equilibrium and optimal processing efficiency amidst diverse and often unpredictable input signals. In response to these challenges, we propose the Asynchronous Bioplausible Neuron (ABN), a dynamic spike firing mechanism to auto-adjust the variations in the input signal. Comprehensive evaluation across various datasets demonstrates ABN's enhanced performance in image classification and segmentation, maintenance of neural equilibrium, and energy efficiency.

Asynchronous Bioplausible Neuron for SNN for Event Vision

TL;DR

The paper tackles the challenge of maintaining homeostasis and energy efficiency in spiking neural networks for event-based vision. It introduces the Asynchronous Bioplausible Neuron (ABN), a per-neuron dynamic-threshold mechanism combining Membrane Gradient, Threshold Retrospective Gradient, and Spike Efficiency, integrated into a spiking MLP with STDP. Across six diverse datasets, ABN delivers state-of-the-art classification and segmentation accuracy, robust performance under degraded inputs, strong homeostasis, and lower energy consumption compared to prior methods. The findings highlight ABN's potential for real-time, neuromorphic vision systems, while acknowledging hardware deployment considerations and opportunities for further biophysiological realism.

Abstract

Spiking Neural Networks (SNNs) offer a biologically inspired approach to computer vision that can lead to more efficient processing of visual data with reduced energy consumption. However, maintaining homeostasis within these networks is challenging, as it requires continuous adjustment of neural responses to preserve equilibrium and optimal processing efficiency amidst diverse and often unpredictable input signals. In response to these challenges, we propose the Asynchronous Bioplausible Neuron (ABN), a dynamic spike firing mechanism to auto-adjust the variations in the input signal. Comprehensive evaluation across various datasets demonstrates ABN's enhanced performance in image classification and segmentation, maintenance of neural equilibrium, and energy efficiency.
Paper Structure (21 sections, 10 equations, 3 figures, 5 tables)

This paper contains 21 sections, 10 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Computer vision applications of SNNs
  4. Dynamic Thresholding in SNNs
  5. Methodology
  6. Pre-requisite
  7. Event-based cameras
  8. Leaky integrate-and-fire
  9. Spike response model
  10. Asynchronous events to spike conversion
  11. Asynchronous Bioplausible Neuron
  12. Membrane Gradient (MG)
  13. Threshold Retrospective Gradient (TRG)
  14. Spike Efficiency (SE)
  15. Results
  16. ...and 6 more sections

Figures (3)

  • Figure 1: Comparative Visualization of SNNs approaches: The top image presents the overview of the SNN methodology showcasing the conventional processes of asynchronous event capture, event-to-frame conversion, and fixed thresholding in neuronal spike response. In contrast, the bottom image illustrates an overview of the proposed method, highlighting the explicit encoding of events to spikes and implementing a dynamic thresholding mechanism for neuronal spiking. This juxtaposition underscores the novel methodology proposed.
  • Figure 2: Dynamic thresholding for SNNs: Existing methods are based on the LIF neuron which uses statistical cues for thresholding (a), however they result in unstable thresholds (b) and consequently low accuracy for segmentation methods (c). Our proposed method incorporates weighted inputs in the form of MG, TRG, and SE (d), which results in a smoothed dynamic threshold (e), and improved segmentation (f).
  • Figure 3: Schematic Diagram of ABN. The DVS camera outputs data in the form of asynchronous events. It is passed into a SNN with neurons in the first layer equal to the number of pixels on the camera. Each neuron adopts the proposed ABN where it takes the weighted input from the MG, TRG, and SE to control the dynamic threshold.