Abstract:Humans learn multiple tasks in succession with minimal mutual interference, through the context gating mechanism in the prefrontal cortex (PFC). The brain-inspired models of spiking neural networks (SNN) have drawn massive attention for their energy efficiency and biological plausibility. To overcome catastrophic forgetting when learning multiple tasks in sequence, current SNN models for lifelong learning focus on memory reserving or regularization-based modification, while lacking SNN to replicate human experimental behavior. Inspired by biological context-dependent gating mechanisms found in PFC, we propose SNN with context gating trained by the local plasticity rule (CG-SNN) for lifelong learning. The iterative training between global and local plasticity for task units is designed to strengthen the connections between task neurons and hidden neurons and preserve the multi-task relevant information. The experiments show that the proposed model is effective in maintaining the past learning experience and has better task-selectivity than other methods during lifelong learning. Our results provide new insights that the CG-SNN model can extend context gating with good scalability on different SNN architectures with different spike-firing mechanisms. Thus, our models have good potential for parallel implementation on neuromorphic hardware and model human's behavior.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the catastrophic forgetting problem faced by Spiking Neural Networks (SNNs) when continuously learning multiple tasks. Specifically, when SNNs learn multiple tasks sequentially, newly learned tasks will interfere with or overwrite the learning results of previous tasks, causing a sharp decline in the network's performance on old tasks. To solve this problem, the paper proposes a Context Gating in Spiking Neural Networks (CG - SNN) model that combines local and global plasticity. ### Main contributions of the paper: 1. **Development of the CG - SNN framework**: - Combine local and global plasticity, and mitigate catastrophic forgetting through the context - gating mechanism induced by local plasticity. - Do not rely on manual weight updates or already - learned samples, but dynamically adjust input information through context gating. 2. **Implementation of single - spike and multi - spike CG - SNN models**: - The single - spike CG - SNN model is implemented by combining global back - propagation plasticity and the local Spike - Timing - Dependent Plasticity (STDP) learning rule. - The multi - spike CG - SNN model is implemented by combining global back - propagation plasticity and the local Oja learning rule. - During the iterative training process, the interaction between global and local plasticity aims to strengthen the connections between task neurons and hidden neurons, thereby retaining multi - task - related information. 3. **Experimental verification**: - The experiment analyzes the effectiveness of context gating in SNN lifelong learning. - The proposed model is compared with the cognitive experimental features in human behavior experiments, and the results show that they are consistent to a certain extent. - The proposed model shows better task selectivity in neural dynamics and has high biological plausibility. ### Core methods of the paper: - **Network architecture**: - A three - layer SNN model, in which the main part is used to receive tree - picture stimuli and output results, and the context signal is input into the network after one - hot label encoding, affecting the output judgment of the network. - Extract and maintain the context information of different tasks by iteratively training with alternating use of global and local plasticity. - **Single - spike CG - SNN**: - Each spiking neuron is only allowed to fire one spike, using the non - leaky Integrate - and - Fire (IF) neuron model. - Membrane potential calculation formula: \[ V_j(t)=\sum_{i = 1}^{N_I}g(t - t_i)w_{ij}(1 - e^{-(t - t_i)}) \] - Use the STDP learning rule for local plasticity training, and the formula is as follows: \[ \Delta W^l_{ij}=\begin{cases} A_+e^{-(t_j - t_i)/\tau_+},&\text{if }t_j > t_i\\ A_-e^{-(t_i - t_j)/\tau_-},&\text{if }t_j < t_i \end{cases} \] - **Multi - spike CG - SNN**: - Global training is based on the Spatio - Temporal Back - Propagation (STBP) algorithm. - Use the Oja learning rule for local plasticity training, and the formula is as follows: \[ w^l_{ij}=w^l_{ij}+\eta_{\text{hebb}}(x - w^l_{ij}y) \] - Introduce "sluggish" neurons to simulate the learning bias of humans in interleaved training, and process the input information of task units through exponential moving average. ### Experimental results: - **Single - spike CG - SNN**: - After applying the STDP learning rule, the performance of the first task declines significantly.

Context Gating in Spiking Neural Networks: Achieving Lifelong Learning through Integration of Local and Global Plasticity

Unleashing the Potential of Spiking Neural Networks for Sequential Modeling with Contextual Embedding.

Adaptive Deep Spiking Neural Network with Global-Local Learning Via Balanced Excitatory and Inhibitory Mechanism

Efficient Spiking Neural Networks with Sparse Selective Activation for Continual Learning

Effect Exploration of the Spiking Neuron Property and Heterogeneity on Brain-Inspired Context-Dependent Learning

Similarity-based context aware continual learning for spiking neural networks

Alleviating catastrophic forgetting using context-dependent gating and synaptic stabilization

Modelling continual learning in humans with Hebbian context gating and exponentially decaying task signals

Toward Efficient Processing and Learning with Spikes: New Approaches for Multispike Learning

Controlled Forgetting: Targeted Stimulation and Dopaminergic Plasticity Modulation for Unsupervised Lifelong Learning in Spiking Neural Networks

A Brain-Inspired Spiking Neural Network Model with Temporal Encoding and Learning

Rapid context inference in a thalamocortical model using recurrent neural networks

Learning the Plasticity: Plasticity-Driven Learning Framework in Spiking Neural Networks

Investigation on the Effect of Spiking Neuron Properties with Dendrite Compartments in Brain-inspired Context-dependent Learning

Brain-inspired Global-Local Learning Incorporated with Neuromorphic Computing

Adaptive Reorganization of Neural Pathways for Continual Learning with Spiking Neural Networks

Enhancing Efficient Continual Learning with Dynamic Structure Development of Spiking Neural Networks

Improving Multispike Learning with Plastic Synaptic Delays

Adaptive Synaptic Scaling in Spiking Networks for Continual Learning and Enhanced Robustness

Spiking generative networks empowered by multiple dynamic experts for lifelong learning

PC-SNN: Supervised Learning with Local Hebbian Synaptic Plasticity Based on Predictive Coding in Spiking Neural Networks