Adaptive Work Budgeting
Cross-source consensus on Adaptive Work Budgeting from 1 sources and 5 claims.
1 sources · 5 claims
How it works
Risks & contraindications
Highlighted claims
- FedQueue sets each client's job budget by subtracting the predicted queue delay and a safety buffer from the synchronization interval. — FedQueue: Queue-Aware Federated Learning for Cross-Facility HPC Training
- Predicted longer queues reduce local work, while predicted shorter queues permit more local work. — FedQueue: Queue-Aware Federated Learning for Cross-Facility HPC Training
- FedQueue converts the time budget into a local step budget using profiled client throughput. — FedQueue: Queue-Aware Federated Learning for Cross-Facility HPC Training
- The framework assumes facility throughput can be profiled and local training respects the assigned job-time budget. — FedQueue: Queue-Aware Federated Learning for Cross-Facility HPC Training
- Safety-buffer size mediates a trade-off between stability and speed. — FedQueue: Queue-Aware Federated Learning for Cross-Facility HPC Training