IBMFL 1.0.1

Release 1.0.1

Major Features

• Supports various types of machine learning models to be trained in a federated learning fashion.
• Multiple state-of-the-art fusion algorithms that fit a variety of federated learning use cases.
• Flask based connection module for easy communication between parties and aggregator.
• Powerful configuration capability for easy testing and switching fusion handlers.
• Extensive examples to demonstrate how to use IBM FL.
• Step-by-step quickstart tutorial for IBM FL.

Highlights

• Supported machine learning model types:
  • Neural networks (any neural network topology supported by Keras)
  • Decision Tree ID3
  • Linear classifiers/regressions (with regularizer): logistic regression, linear SVM, ridge regression, KMeans, and Naïve Bayes
  • Deep Reinforcement Learning algorithms including DQN, DDPG, PPO and more
• Supported fusion algorithms:
  • Iterative Average
  • FedAvg McMahan et al.
  • Gradient Average
  • Probabilistic Federated Neural Matching (PFNM) Yurochkin et al.
  • Krum Blanchard et al.
  • Coordinated median Yin et al.
  • Zeno Xie et al.
  • Statistical Parameter Aggregation via Heterogeneous Matching (SPAHM) Yurochkin et al.
  • ID3 fusion for decision tree Quinlan
  • Naive Bayes fusion with differential privacy