🌐 Federated Learning Core

FedAvg • Non-IID Partitioning • Compression • Personalization • DP-SGD

Overview • About • Topics • Quick Start • Experiments

---

Core federated learning repository covering algorithm baselines, partitioning strategies, communication efficiency, personalization, and privacy-aware training.

---

🎯 Overview

federated-learning-core provides reusable FL building blocks:

• FedAvg/FedProx/FedAdam orchestration layers
• IID and non-IID client data partitioning
• Compression strategies for communication cost reduction
• Personalization modules for client-specific adaptation
• DP-SGD privacy mechanisms and accounting

📌 About

• Engineered for experimentation and modular FL system design
• Supports research workflows and production migration paths
• Includes ready-to-run scenario and experiment scripts

🏷️ Topics

federated-learning fedavg non-iid flower differential-privacy distributed-ml pytorch privacy

🧩 Architecture

• src/algorithms/: aggregation and strategy modules
• src/data/partitioning/: IID and skew partitioners
• src/training/flower/: orchestrated FL training
• src/optimization/: compression and personalization
• src/privacy/dp_sgd/: private training components
• src/scenarios/: cross-silo and vertical setups

⚡ Quick Start

pip install -r requirements.txt
pytest -q tests/test_public_surfaces.py

🧪 Experiments

• src/experiments/run_fedavg.py
• src/experiments/run_flower.py
• src/experiments/run_compression.py
• src/experiments/run_cross_silo.py
• src/experiments/run_vertical.py
• src/experiments/run_personalization.py
• src/experiments/run_dp.py

🛠️ Tech Stack

FL/ML: PyTorch, Flower, NumPy, scikit-learn
Optimization: compression + personalization techniques
Privacy: DP-SGD and accounting utilities