Modeling of a Hedonic Synapse in a LIF Neuron and its Application to Spiking Neural Networks
Modeling of a Hedonic Synapse in a LIF Neuron and its Application to Spiking Neural Networks
Azaryan Sergey, Otaryan Knar
Summary
Key words: (LIF), Reward-Modulated Plasticity, Hedonic Learning, Surrogate Gradient, Reinforcement Learning, Temporal Signal Processing, Numerical Simulation
This work investigates hedonic synapses as a mechanism for global learning in spiking neural networks. A Leaky Integrate-and-Fire (LIF) neuron model with reward-modulated synaptic plasticity is proposed, where synaptic weights are updated using a global reinforcement signal. The approach complements local plasticity rules such as STDP and supports integration into hybrid SNN–RL architectures. The model is implemented in Python 3.12 using a Euler-based numerical scheme. Learning is driven by a reward/penalty signal defined as the difference between target and actual spike counts. Simulation results demonstrate numerical stability, bounded weight dynamics, and effective adaptation of synaptic parameters. The findings confirm that hedonic learning is a promising framework for temporal signal processing and biologically inspired learning systems. The model provides high interpretability and can serve as a flexible testbed for studying global plasticity. Future work may include scaling to larger networks and analyzing the effects of noise and stochasticity.