Show HN: A 4-year-old "TurboQuant" implementation

Category: library

Tags: federated-learning, distributed-mean-estimation, gradient-compression

Score: 5.3/10 (Innovation: 6, Technical: 7, Documentation: 4, Utility: 4)

EDEN is a communication-efficient and robust distributed mean estimation technique for federated learning, presented at ICML 2022. It handles heterogeneous communication budgets and packet losses by offering a lossy unbiased compression method with theoretical guarantees. The project provides implementations in PyTorch and TensorFlow but lacks detailed usage examples and active maintenance.

Target audience: machine-learning-engineers

Repository: https://github.com/amitport/EDEN-Distributed-Mean-Estimation · Jupyter Notebook · 14 stars

View on Hacker News