Abstract:By exploiting discrete signal processing and simulating brain neuron communication, Spiking Neural Networks (SNNs) offer a low-energy alternative to Artificial Neural Networks (ANNs). However, existing SNN models, still face high computational costs due to the numerous time steps as well as network depth and scale. The tens of billions of neurons and trillions of synapses in the human brain are developed from only 20,000 genes, which inspires us to design an efficient genetic encoding strategy that dynamic evolves to regulate large-scale deep SNNs at low cost. Therefore, we first propose a genetically scaled SNN encoding scheme that incorporates globally shared genetic interactions to indirectly optimize neuronal encoding instead of weight, which obviously brings about reductions in parameters and energy consumption. Then, a spatio-temporal evolutionary framework is designed to optimize the inherently initial wiring rules. Two dynamic regularization operators in the fitness function evolve the neuronal encoding to a suitable distribution and enhance information quality of the genetic interaction respectively, substantially accelerating evolutionary speed and improving efficiency. Experiments show that our approach compresses parameters by approximately 50\% to 80\%, while outperforming models on the same architectures by 0.21\% to 4.38\% on CIFAR-10, CIFAR-100 and ImageNet. In summary, the consistent trends of the proposed genetically encoded spatio-temporal evolution across different datasets and architectures highlight its significant enhancements in terms of efficiency, broad scalability and robustness, demonstrating the advantages of the brain-inspired evolutionary genetic coding for SNN optimization.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to address the high computational cost issue of Spiking Neural Networks (SNNs). Although SNNs offer a low-energy computation paradigm by simulating the communication characteristics of brain neurons, existing SNN models still face high computational costs due to a large number of time steps, network depth, and scale. Specifically, the paper raises the following issues: 1. **High computational cost**: Existing SNN models have high computational costs due to many time steps, network depth, and large scale. 2. **Large number of parameters**: SNN models typically have a large number of parameters, increasing the burden of storage and computation. 3. **Limited optimization methods**: Current SNN optimization methods mainly focus on training mechanisms, such as plasticity-based training, conversion-based training, and gradient-based training, but these methods have limited effectiveness in reducing computational costs. To address these issues, the paper proposes a genetic encoding evolutionary strategy by designing an efficient genetic encoding scheme and a spatiotemporal evolutionary framework to optimize the computational efficiency and performance of large-scale deep SNNs. The specific goals include: - **Reducing the number of parameters**: Significantly reduce the number of network parameters through genetic encoding methods, thereby lowering storage and computational costs. - **Improving energy efficiency**: Accelerate the evolution speed and improve energy efficiency by optimizing initial connection rules and dynamic regularization operations. - **Enhancing performance**: Achieve a significant reduction in the number of parameters while maintaining or improving performance. Experimental results show that this method not only significantly reduces the number of parameters but also outperforms other models with the same architecture in terms of performance on the CIFAR-10, CIFAR-100, and ImageNet datasets.

Evolving Efficient Genetic Encoding for Deep Spiking Neural Networks

Efficient Structure Slimming for Spiking Neural Networks

Hierarchical Spiking-Based Model for Efficient Image Classification with Enhanced Feature Extraction and Encoding.

Deep CovDenseSNN: A Hierarchical Event-Driven Dynamic Framework with Spiking Neurons in Noisy Environment

Brain-inspired Evolutionary Architectures for Spiking Neural Networks

Spike Trains Encoding and Threshold Rescaling Method for Deep Spiking Neural Networks

An Efficient Learning Algorithm for Direct Training Deep Spiking Neural Networks

Adaptive Spiking Neural Networks with Hybrid Coding

Spiking Deep Residual Networks.

Multi-scale Evolutionary Neural Architecture Search for Deep Spiking Neural Networks

High-performance deep spiking neural networks via at-most-two-spike exponential coding

Neurogenesis Dynamics-inspired Spiking Neural Network Training Acceleration

You Only Spike Once: Improving Energy-Efficient Neuromorphic Inference to ANN-Level Accuracy

Training Energy-Efficient Deep Spiking Neural Networks with Single-Spike Hybrid Input Encoding

ESL-SNNs: An Evolutionary Structure Learning Strategy for Spiking Neural Networks

30.2 A 22nm 0.26nW/Synapse Spike-Driven Spiking Neural Network Processing Unit Using Time-Step-First Dataflow and Sparsity-Adaptive In-Memory Computing

Exploring Optimized Spiking Neural Network Architectures for Classification Tasks on Embedded Platforms

Scaling Spike-driven Transformer with Efficient Spike Firing Approximation Training

Stochastic Spiking Neural Networks with First-to-Spike Coding

Gated Attention Coding for Training High-performance and Efficient Spiking Neural Networks

Canonic Signed Spike Coding for Efficient Spiking Neural Networks