HASA: Subnet Allocation for Compute-Constrained Model-Heterogeneous Federated Learning

📰 ArXiv cs.AI

Learn how HASA allocates subnets for model-heterogeneous federated learning to reduce client costs and improve performance in compute-constrained environments

advanced Published 9 Jun 2026

Action Steps

Implement HASA to allocate subnets for model-heterogeneous federated learning
Train a shared supernet and allocate subnets to clients based on their compute resources
Evaluate the performance of HASA against other subnet-allocation policies
Apply HASA to real-world federated learning deployments to reduce client costs
Compare the results of HASA with other optimization techniques for federated learning

Who Needs to Know This

Machine learning engineers and researchers working on federated learning projects can benefit from this article to optimize their model training and deployment

Key Insight

💡 HASA explicitly accounts for device constraints and local data distributions to optimize subnet allocation

Key Takeaways

Learn how HASA allocates subnets for model-heterogeneous federated learning to reduce client costs and improve performance in compute-constrained environments

Full Article

Title: HASA: Subnet Allocation for Compute-Constrained Model-Heterogeneous Federated Learning

Abstract:
arXiv:2606.07621v1 Announce Type: cross Abstract: Edge services increasingly use federated learning to personalize on-device models while keeping sensitive data local. In practice, deployments must handle heterogeneity in both client resources and local data distributions. Model-heterogeneous federated learning lowers client cost by allowing each client to train a subnet of a shared supernet, but most subnet-allocation policies are driven by device constraints and do not explicitly account for s

Read full paper → ← Back to Reads