Federated Machine Learning

Warning

This post is more than a year old. Information may be outdated.

text

Repeat For each client (in parallel):
    1. Receive global model (R1 - R3)
    2. Train global model on a local dataset
    3. Reply local model (R4 - R6)
    Aggregate local models into a new global model
For T rounds

Federated learning is a privacy-preserving machine learning approach that enables multiple parties to train a machine learning model collaboratively without sharing their data with a central server. The goal is to allow the parties to jointly learn a model representative of their collective data while preserving the privacy of individual data points. In federated learning, each party trains a local model on their data then sends updates of the model's parameters to a central server, aggregating the updates to create a new global model. The process is repeated iteratively until the global model converges to a satisfactory level of accuracy.

To preserve privacy, differential privacy adds noise to the updates before they are sent to the central server. This makes it difficult for any party to learn anything about the data of the other parties from the updates while still allowing the server to aggregate the updates and learn from the collective data.

Backlinks8

Federated Transfer Learning Horizontal Federated Learning Vertical Federated Learning honest-but-curious Indirect Information Leakage Qiang Yang et al. Federated Machine Learning Concept and Applications Tian Li et al. Federated Learning Challenges, Methods, and Future Directions 230419

Comments0

No comments yet. Be the first to share your thoughts.

Comments