Impact of Internal Noise on the Performance of Trained Spiking Neural Networks
Daniil A. Maksimov1, Victor M. Moskvitin1, Nadezhda I. Semenova1; 1Saratov State University, Saratov, Russia;
Abstract
This study examines the impact of additive and multiplicative noise on a trained spiking neural network (SNN). Noise was introduced at various stages of signal processing in the neural network, including the input current, membrane potential, and spike output generation.
The results show that multiplicative noise applied to the membrane potential has the greatest negative impact on network performance, leading to a significant reduction in accuracy. This is primarily due to its tendency to suppress membrane potentials to large negative values, effectively silencing neural activity.
To address this issue, strategies for pre‑filtering inputs were evaluated, and the sigmoid activation function filter demonstrated the best performance, shifting the input data into a strictly positive range. Under such conditions, additive noise in the input current becomes the dominant source of performance degradation, whereas other noise configurations reduce accuracy by no more than 1 %, even at high noise intensity.
Speaker
Maksimov Daniil Alekseevich
Saratov State University, Saratov, Russia
Russia
Discussion
Ask question
