Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Izhikevich-Inspired Temporal Dynamics for Enhancing Privacy, Efficiency, and Transferability in Spiking Neural Networks

Ayana Moshruba, Hamed Poursiami, Maryam Parsa

TL;DR

This work investigates how biologically inspired temporal spike dynamics can enhance privacy, efficiency, and transferability in scalable SNNs. By introducing two input-level mechanisms, Poisson-Burst and Delayed-Burst, the authors emulate burst variability and latency-driven spiking while keeping Leaky Integrate-and-Fire neurons intact. Across image and tabular datasets, these dynamics yield improved resilience to membership inference attacks, with Poisson-Burst maintaining near-baseline accuracy and Delayed-Burst offering stronger privacy at a cost to accuracy. Transfer learning experiments show that temporal diversity can bolster generalization and privacy under domain shifts, suggesting practical benefits for privacy-aware neuromorphic systems and energy-efficient hardware implementations.

Abstract

Biological neurons exhibit diverse temporal spike patterns, which are believed to support efficient, robust, and adaptive neural information processing. While models such as Izhikevich can replicate a wide range of these firing dynamics, their complexity poses challenges for directly integrating them into scalable spiking neural networks (SNN) training pipelines. In this work, we propose two probabilistically driven, input-level temporal spike transformations: Poisson-Burst and Delayed-Burst that introduce biologically inspired temporal variability directly into standard Leaky Integrate-and-Fire (LIF) neurons. This enables scalable training and systematic evaluation of how spike timing dynamics affect privacy, generalization, and learning performance. Poisson-Burst modulates burst occurrence based on input intensity, while Delayed-Burst encodes input strength through burst onset timing. Through extensive experiments across multiple benchmarks, we demonstrate that Poisson-Burst maintains competitive accuracy and lower resource overhead while exhibiting enhanced privacy robustness against membership inference attacks, whereas Delayed-Burst provides stronger privacy protection at a modest accuracy trade-off. These findings highlight the potential of biologically grounded temporal spike dynamics in improving the privacy, generalization and biological plausibility of neuromorphic learning systems.

Izhikevich-Inspired Temporal Dynamics for Enhancing Privacy, Efficiency, and Transferability in Spiking Neural Networks

TL;DR

This work investigates how biologically inspired temporal spike dynamics can enhance privacy, efficiency, and transferability in scalable SNNs. By introducing two input-level mechanisms, Poisson-Burst and Delayed-Burst, the authors emulate burst variability and latency-driven spiking while keeping Leaky Integrate-and-Fire neurons intact. Across image and tabular datasets, these dynamics yield improved resilience to membership inference attacks, with Poisson-Burst maintaining near-baseline accuracy and Delayed-Burst offering stronger privacy at a cost to accuracy. Transfer learning experiments show that temporal diversity can bolster generalization and privacy under domain shifts, suggesting practical benefits for privacy-aware neuromorphic systems and energy-efficient hardware implementations.

Abstract

Biological neurons exhibit diverse temporal spike patterns, which are believed to support efficient, robust, and adaptive neural information processing. While models such as Izhikevich can replicate a wide range of these firing dynamics, their complexity poses challenges for directly integrating them into scalable spiking neural networks (SNN) training pipelines. In this work, we propose two probabilistically driven, input-level temporal spike transformations: Poisson-Burst and Delayed-Burst that introduce biologically inspired temporal variability directly into standard Leaky Integrate-and-Fire (LIF) neurons. This enables scalable training and systematic evaluation of how spike timing dynamics affect privacy, generalization, and learning performance. Poisson-Burst modulates burst occurrence based on input intensity, while Delayed-Burst encodes input strength through burst onset timing. Through extensive experiments across multiple benchmarks, we demonstrate that Poisson-Burst maintains competitive accuracy and lower resource overhead while exhibiting enhanced privacy robustness against membership inference attacks, whereas Delayed-Burst provides stronger privacy protection at a modest accuracy trade-off. These findings highlight the potential of biologically grounded temporal spike dynamics in improving the privacy, generalization and biological plausibility of neuromorphic learning systems.
Paper Structure (19 sections, 3 equations, 5 figures, 4 tables)

This paper contains 19 sections, 3 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Background
  4. Temporal Dynamics in SNNs
  5. Poisson-Burst Temporal Dynamics
  6. Delayed-Burst Temporal Dynamics
  7. Membership Inference Attack (MIA)
  8. Transfer Learning
  9. Experimental Setup
  10. Dataset and Model Architecture
  11. Temporal Dynamics Implementation
  12. Poisson-Burst Dynamics
  13. Delayed-Burst Dynamics
  14. Results
  15. Model Performance
  16. ...and 4 more sections

Figures (5)

  • Figure 1: Illustration of two biologically inspired temporal spike dynamics introduced in this study. (Left) Poisson-Burst encoding generates bursts with variable spike counts sampled from a Poisson distribution, modeling repetitive firing. (Right) Delayed-Burst encoding introduces latency-based spike onset, where higher input intensities lead to earlier fixed-length bursts. Each row represents increasing input intensity from top to bottom.
  • Figure 2: Membership Inference Attack (MIA) Framework
  • Figure 3: Impact of different temporal spike dynamics on model performance across (a) Fashion-MNIST, (b) CIFAR-10 and (c) Iris datasets.
  • Figure 4: ROC curves illustrating the effect of different temporal spike dynamics on MIA resilience across (a) FMNIST, (b) CIFAR-10, and (c) Iris datasets.
  • Figure 5: Computational cost on MNIST across temporal spike dynamics: (a) GPU power, (b) GPU memory, (c) CPU memory.