Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improvement of Spiking Neural Network with Bit Planes and Color Models

Nhan T. Luu, Duong T. Luu, Nam N. Pham, Thang C. Truong

TL;DR

This work tackles the challenge of improving SNN performance on image tasks by introducing a dual input encoding scheme that combines traditional spike coding with bit-plane decomposition of color-converted inputs. A theoretical framework based on surrogate-gradient training and gradient-SNR supports the proposed bit-plane concatenation, and extensive experiments show consistent accuracy gains across grayscale and color datasets, with particularly strong improvements on color images. The study also analyzes the impact of different color models, finding that while RGB is not always optimal, space choices like HSL can yield higher gains at the cost of increased computation. The findings point to a viable path for more accurate and potentially energy-efficient SNNs and suggest practical avenues for hardware implementations in neuromorphic systems.

Abstract

Spiking neural network (SNN) has emerged as a promising paradigm in computational neuroscience and artificial intelligence, offering advantages such as low energy consumption and small memory footprint. However, their practical adoption is constrained by several challenges, prominently among them being performance optimization. In this study, we present a novel approach to enhance the performance of SNN for images through a new coding method that exploits bit plane representation. Our proposed technique is designed to improve the accuracy of SNN without increasing model size. Also, we investigate the impacts of color models of the proposed coding process. Through extensive experimental validation, we demonstrate the effectiveness of our coding strategy in achieving performance gain across multiple datasets. To the best of our knowledge, this is the first research that considers bit planes and color models in the context of SNN. By leveraging the unique characteristics of bit planes, we hope to unlock new potentials in SNNs performance, potentially paving the way for more efficient and effective SNNs models in future researches and applications.

Improvement of Spiking Neural Network with Bit Planes and Color Models

TL;DR

This work tackles the challenge of improving SNN performance on image tasks by introducing a dual input encoding scheme that combines traditional spike coding with bit-plane decomposition of color-converted inputs. A theoretical framework based on surrogate-gradient training and gradient-SNR supports the proposed bit-plane concatenation, and extensive experiments show consistent accuracy gains across grayscale and color datasets, with particularly strong improvements on color images. The study also analyzes the impact of different color models, finding that while RGB is not always optimal, space choices like HSL can yield higher gains at the cost of increased computation. The findings point to a viable path for more accurate and potentially energy-efficient SNNs and suggest practical avenues for hardware implementations in neuromorphic systems.

Abstract

Spiking neural network (SNN) has emerged as a promising paradigm in computational neuroscience and artificial intelligence, offering advantages such as low energy consumption and small memory footprint. However, their practical adoption is constrained by several challenges, prominently among them being performance optimization. In this study, we present a novel approach to enhance the performance of SNN for images through a new coding method that exploits bit plane representation. Our proposed technique is designed to improve the accuracy of SNN without increasing model size. Also, we investigate the impacts of color models of the proposed coding process. Through extensive experimental validation, we demonstrate the effectiveness of our coding strategy in achieving performance gain across multiple datasets. To the best of our knowledge, this is the first research that considers bit planes and color models in the context of SNN. By leveraging the unique characteristics of bit planes, we hope to unlock new potentials in SNNs performance, potentially paving the way for more efficient and effective SNNs models in future researches and applications.

Paper Structure

This paper contains 20 sections, 6 theorems, 49 equations, 5 figures, 8 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Spiking Neural Network
  3. Overview of SNNs
  4. Deep Residual Learning in SNNs
  5. Bit Planes and Color Models in Deep Learning
  6. Bit planes in deep learning
  7. Color models in deep learning
  8. Proposed Method
  9. Theoretical basis
  10. Experiments
  11. Experimental settings
  12. Experimental Results
  13. The effect of bit plane coding on grayscale image datasets
  14. Comparison of SNN variants
  15. The effect of bit plane coding on color images
  16. ...and 5 more sections

Key Result

Theorem A.1

Let $\mathbf{x}_{1}^{k}$ denote a $k$-tuple $(\mathbf{x}_{1},\ldots,\mathbf{x}_{k})$ of input instances, and assume that each $\mathbf{x}_{l}$ is i.i.d. standard Gaussian in $\mathbb{R}^{d}$. Define and the objective (w.r.t. some predictor $p_{\mathbf{w}}$ parameterized by $\mathbf{w}$) Where the loss function $\ell$ is either the square loss $\ell(\hat{y},y)=\frac{1}{2}(\hat{y}-y)^{2}$ or a cla

Figures (5)

  • Figure 1: Visualization of the extracted bit planes from a sample image in the CIFAR-10 dataset alex2009learning. Higher-order bit planes (e.g., most significant bits) encode more spatial and structural information, capturing the prominent features of the image. In contrast, lower-order bit planes (e.g., least significant bits) primarily contain finer details and noise, contributing less to the overall image structure.
  • Figure 2: Our proposed coding method (lower) compare with traditional spike coding method (upper).
  • Figure 3: Comparing model gradient's SNR of the proposed method (red), baseline (blue) and bit plane coding (green) of MNIST dataset across the epochs. Dashed line indicate the mean gradient SNR (higher is better based on our assumption).
  • Figure 4: Validation accuracy visualization of proposed method among color models.
  • Figure 5: Relationship between average gain in performance (higher is better) and average increase in processing time (lower is better) of different color models.

Theorems & Definitions (9)

  • Theorem A.1: Shalev et al., 2017
  • Lemma A.2: Shalev et al., 2017
  • Lemma A.3: Shalev et al., 2017
  • Theorem B.1
  • proof
  • Lemma B.2
  • proof
  • Lemma B.3
  • proof