TY  - JOUR
T1  - Analysis on Performance and Energy Consumption in Image Classification with SNN-Based Federated Learning
AU  - Kim, Dong-gyun 
AU  - Lim, Jae-han 
AU  - Lee, Hyun-jong 
AU  - Lim, Yeon-sup 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2024
DA  - 2024/1/1
DO  - 10.7840/kics.2024.49.7.935
KW  - Federated Learning
KW  - Spiking Neural Network
KW  - 2-dimensional image classification
AB  - Federated learning is a cooperative learning process of global neural network models with several devices.
Many researchers focus on it due to its advantages in data privacy protection and low communication costs.
Most previous studies on federated learning use conventional Artificial Neural Networks (ANNs) as global model. However, it is difficult to use ANNs in mobile and embedded devices because they consume a lot of energy. To apply federate learning to mobile and embedded devices successfully, it is important to use Spiking Neural Networks (SNNs) as global model due to their high energy efficiency. This is because SNNs deliver data between neurons with spikes and the delivery process works in an event-driven manner, which is more energy efficient than that of ANNs. However, the studies on federated learning with SNNs are much less than the studies with ANNs. This problem makes the federated learning with SNNs difficult to be utilized in various applications. In this paper, we conduct simulation with diverse situations and find out that the federated learning with SNNs is more energy efficient than the federated learning with ANNs.