adap/flower

Features

• Federated Orchestration - Functions as a coordinator managing model aggregation and communication between a central server and decentralized clients.
• Privacy-Preserving Model Training - Enables the training of machine learning models on decentralized hardware to keep sensitive data local while improving a global model.
• Model Aggregation - Coordinates the collection and combination of local model updates into a synchronized global model.
• Federated Aggregation Strategies - Employs interchangeable strategy classes to determine how local model updates are aggregated into a global model.
• Local Model Training Integrations - Provides a framework for executing model training and weight updates on decentralized local hardware.
• Federated Learning Frameworks - Functions as a comprehensive framework for orchestrating collaborative machine learning and analytics across decentralized data sources.
• Custom Federated Strategies - Provides the ability to implement and manage custom coordination strategies for the collection and distribution of model updates.
• ML Library Integrations - Integrates with a wide variety of machine learning libraries to enable federated training across diverse frameworks.
• Model Aggregation Strategies - Implements interchangeable strategies for aggregating local model updates from remote workers into a global model.
• Privacy-Preserving Training - Enables training of AI models on local hardware to keep sensitive information on-device.
• AI Client-Server Communication Protocols - Implements specialized bidirectional messaging protocols for exchanging model weights and metadata between servers and remote workers.
• Remote Procedure Calls - Uses a remote procedure call system to distribute training tasks and collect results from remote devices.
• Federated Toolkits - Provides a comprehensive toolkit for performing data analysis and model training on decentralized hardware.
• Distributed Training Coordination - Coordinates the synchronization and management of machine learning training tasks across a network of distributed devices.
• ML Framework Abstractions - Provides an agnostic wrapper that allows different machine learning libraries to operate within the same training loop.
• ML Framework Abstraction Layers - Employs an agnostic wrapper to connect diverse machine learning libraries within the same training loop.
• Weight Serialization - Manages the serialization of model weights for efficient transmission across the network between coordinators and workers.
• Decentralized Data Analysis - Performs data analysis across separate decentralized datasets without requiring the raw information to be centralized.
• Distributed Task Orchestration - Orchestrates the distribution and collection of training tasks across a network of devices using a structured request-response pattern.
• Runtime Coordination - Coordinates the execution state and lifecycle of local training processes across multiple independent decentralized nodes.
• Model Weight Exchange Protocols - Uses a remote procedure call orchestrator to manage the exchange of model weights and metadata.
• Federated Workflow Message Buses - Provides a custom message bus to transmit non-standard data packets for specialized federated learning workflows.
• Federated Learning - Friendly framework for implementing federated learning.
• Model Implementations - Example of federated learning integration.
• Privacy and Safety - Unified framework for federated learning across workloads.

Star history