← Federated Learning

Federated Learning — Section

Federated learning — environmental sustainability

Carbon footprint monitoring and green AI practices for federated learning.

Environmental sustainability

Carbon footprint monitoring

Measurement tools:

  • CodeCarbon: Python package for tracking CO₂ emissions [1].

    from codecarbon import EmissionsTracker
tracker = EmissionsTracker()
tracker.start()
# Run FL training
emissions = tracker.stop()
  • ML CO2 Impact: Online calculator for ML carbon footprint.
  • Green Algorithms: Computational footprint calculator.

Optimisation strategies:

  • Schedule training during low-carbon energy periods.
  • Use model compression techniques (pruning, quantization).
  • Implement early stopping based on carbon budget.
  • Prefer edge devices over cloud GPUs when possible.
  • Adaptive client‑selection (EcoLearn) cuts CO₂ by up to ten × without accuracy loss [2].

See: EcoFL framework (arXiv 2023).

Bibliography

  1. Courty, Benoit, Schmidt, Victor, Luccioni, Sasha, et al. (2024). mlco2/codecarbon: v2.4.1. DOI: 10.5281/zenodo.11171501

  2. Mehboob, Talha, Bashir, Noman, Iglesias, Jesus Omana, Zink, Michael, Irwin, David (2023). CEFL: Carbon-efficient federated learning. arXiv preprint arXiv:2310.17972. Available at: https://arxiv.org/abs/2310.17972