scatter_plot InteFL¶
Federated learning execution & research framework
shield 9 Strategies | bug_report 11 Attacks | database 20+ Datasets
What Is InteFL?
InteFL is a full-stack federated learning platform for running, attacking, and defending distributed ML simulations. Configure a JSON file, launch a simulation, and get CSV metrics, PDF plots, and attack snapshot reports — all orchestrated through a React dashboard or CLI.
Configure. Simulate. Analyse.
Define your experiment in a single JSON config. InteFL handles dataset partitioning, client orchestration, adversarial injection, and results collection. Compare strategies side-by-side. Reproduce every run.
Research-grade reproducibility, out of the box.
One Config, Full Pipeline
JSON config → partitioned datasets → N federated clients → aggregated metrics, plots, and snapshots
Explore the Docs
Install with Docker or locally and run your first simulation in minutes
Every intellifl-dev command — setup, dev, test, lint, and more
How the API, Celery workers, Flower engine, and React UI fit together
Full StrategyConfig field reference — every knob you can turn
FEMNIST, FLAIR, MedMNIST, CIFAR-100, HuggingFace text, and more
FedAvg, Krum, Multi-Krum, Bulyan, RFA, PID, Trust, Trimmed Mean, ArKrum
Label flipping, backdoors, model poisoning, Byzantine perturbation, and more
REST endpoints for launching, monitoring, and managing simulations
Key Features
FedAvg, Krum, Multi-Krum, Bulyan, RFA, Trimmed Mean, PID-based, Trust-based, ArKrum — each with configurable parameters
Data poisoning: label flipping, targeted label flipping, Gaussian noise, backdoor triggers, token replacement
Model poisoning: model poisoning, gradient scaling, boosted scaling, Byzantine perturbation, inner product manipulation, alternating min
FEMNIST, FLAIR, CIFAR-100, 11 MedMNIST subsets, Lung Cancer, plus HuggingFace text datasets (financial, legal, medical)
REST API + React dashboard, Celery task queue, SSE live streaming, Docker Compose deployment, and this docs site — all included