کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Guided Training of Spiking Neural Networks via an Auxiliary Artificial Neural Network

نویسندگان :

Neda Falah¹ Majid Shalchian²

1- دانشگاه صنعتی امیرکبیر(پلی تکنیک تهران) 2- دانشگاه صنعتی امیرکبیر(پلی تکنیک تهران)

کلمات کلیدی :

Spiking Neural Networks (SNNs)،Artificial Neural Networks (ANNs)،Guided Learning،Hybrid Training،MNIST Classification

چکیده :

Spiking Neural Networks (SNNs) offer brain-inspired, energy-efficient computation, but their training is hindered by non-differentiable spike signals. This paper introduces a novel hybrid framework where an auxiliary Artificial Neural Network (ANN) guides weight updates for an SNN. The ANN observes the SNN’s internal states—such as membrane potentials and spike activities—and predicts gradient-like updates, enabling stable and efficient optimization. Experiments on MNIST demonstrate that the proposed SNN+ANN model achieves 97.7% accuracy, surpassing the 92% accuracy of a baseline SNN, while also converging faster and with greater stability. The method effectively combines the temporal efficiency of spiking computation with the robust optimization of differentiable networks, offering a promising direction for neuromorphic learning systems.