p07 / Conference
Intelligent client selection for federated learning using cellular automata
Cellular Nanoscale Networks and their Applications (CNNA 2023), 1-4 · 2023
In plain language
CA-CS selects federated clients according to changing device and network conditions instead of choosing participants uniformly at random.
Research context
Client selection can reduce slow-round bottlenecks without requiring every device to participate.
Approach
A cellular-automaton model accounts for computing resources, communication capacity, and interactions between neighboring clients. Evaluation uses MNIST and CIFAR-10.
Findings
CA-CS produces accuracy comparable to random selection while avoiding clients with high latency.
Summary based on abstract.