BackFed: an Efficient and Standardized Benchmark Suite for Backdoor Attacks in Federated Learning

Federated Learning FL systems are vulnerable to backdoor attacks, where adversaries train their local models on poisoned data and submit poisoned model updates to compromise the global model. Despite numerous proposed attacks and defenses, divergent experimental settings, implementation errors, a...