Reproduce “Collate: Collaborative Neural Network Learning for Latency-Critical Edge Systems” 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

Shuo Huai, Di Liu, Hao Kong, Xiangzhong Luo, Weichen Liu, Ravi Subramaniam, Christian Makaya, Qian Lin

Collate is a collaborative neural network learning framework designed for diverse edge systems that enables simultaneous training of custom-architectured deep neural networks for each device. It optimizes for individual inference latency constraints and high accuracy within a single federated learning process, achieving an average accuracy improvement of 1.96% for extended models and 3.09% for shrunk models compared to other methods.

Key insights

  • Optimizing inference latency for heterogeneous edge devices within a federated learning context necessitates dynamic, adaptive model architecture adjustments tailored for each client's specific constraints.
  • Beyond merely shrinking models for resource-constrained devices, leveraging the excess computational capacity of powerful edge devices through model extension can yield tangible accuracy benefits.
  • A prototype-based correction mechanism in global aggregation can effectively mitigate accuracy degradation in heterogeneous federated learning by guiding diverse local models toward a consistent global knowledge representation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.