TALLOP: Joint Optimization of Local Training and Communication Compression for Time-Constrained Federated Learning

Key points

• Nevertheless, in heterogeneous environments, slow clients with limited computational and communication capabilities can delay global synchronization, forcing faster clients to wait and increasing the overall training time.1
• Although previous studies have explored such joint optimization, determining an effective training--communication balance under a fixed per-round time budget remains challenging, as the resulting model performance depends on both training progress and compression-induced degradation.1
• We therefore propose TALLOP, which jointly optimizes the local training workload and communication compression under per-round time constraints.1
• Experiments in heterogeneous FL environments show that TALLOP generally achieves competitive or improved time-to-accuracy compared with approaches that optimize only local training or neither, although the benefit over training-only adaptation depends on the setting.1

Sources (1)

• [1]TALLOP: Joint Optimization of Local Training and Communication Compression for Time-Constrained Federated Learning
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 16, 06:20 AM

  Minimizing the wall-clock time required to reach a target accuracy is crucial in federated learning (FL), where many applications require frequent model updates as data distributions evolve over time.

  Nevertheless, in heterogeneous environments, slow clients with limited computational and communication capabilities can delay global synchronization, forcing faster clients to wait and increasing the overall training time.

Extractive summary: sentences quoted from the sources.