Reproduce “Real-time Federated Evolutionary Neural Architecture Search” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hangyu Zhu, Yaochu Jin
A real-time federated evolutionary neural architecture search (FNAS) framework optimizes neural network performance and efficiency in federated learning. This approach integrates multi-objective evolutionary optimization with a double-sampling technique and weight inheritance, allowing for continuous architecture adaptation and stable performance without model reinitialization.
Key insights
- Integrating evolutionary multi-objective optimization with federated learning can simultaneously optimize network architecture for both performance and efficiency in real-time.
- Double-sampling (model and client) significantly reduces communication and computation overhead, enabling practical, continuous NAS on resource-constrained edge devices.
- Weight inheritance from a global master model ensures stable performance throughout the real-time evolutionary search, preventing drops associated with reinitialization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.