Скачать или смотреть FL Framework – Frontend UX Showcase (Prototype Demo)

FL Framework – UX Showcase

This video presents a quick demo of the FL Framework frontend user experience, designed for student researchers and developers working in the field of Federated Learning.

🎯 Purpose:
To demonstrate a research-focused web interface for configuring, executing, and analyzing federated learning simulations with advanced attack mitigation strategies and dynamic dataset poisoning.

🔧 Tech Stack:
Frontend:
• React 19.1 with Vite 7.1
• React Bootstrap 2.10 (Material Design 3 themed)
• Recharts 3.2 for interactive visualizations
• React Router 7.9 for navigation
• Axios for API communication

Backend:
• Python FastAPI with uvicorn
• Flower 1.9.0 (Federated Learning framework)
• PyTorch 2.2.2 with GPU acceleration support
• Transformers 4.56.1 for LLM support
• HuggingFace Datasets for data loading

Core ML Libraries:
• scikit-learn for clustering & metrics
• pandas & numpy for data processing
• PEFT for LoRA fine-tuning support

🧩 Key Features:
• Interactive Dashboard: Browse and manage simulation runs with real-time status tracking
• Simulation Configuration: Comprehensive form with 30+ parameters including:
HuggingFace dataset integration with real-time validation
Multiple partitioning strategies (IID, Dirichlet, Pathological)
Attack configuration (label flipping, Gaussian noise, brightness attacks)
Strategy selection (FedAvg, Trust, PID, Multi-Krum, Bulyan, RFA, Trimmed-Mean)
• Dynamic Dataset Poisoning: Round-based attack scheduling with configurable ratios
• Training Visualization: Real-time progress monitoring and interactive Recharts plots
• Results Analysis:
Per-client metric visualization with malicious client highlighting
Round-level aggregate metrics
Comparison view for multi-simulation analysis
Zoom & brush controls for detailed inspection
• Dark/Light Theme: Material Design 3 theming with persistent preferences

🔬 Supported Aggregation Strategies:
• FedAvg (baseline)
• Trust & Reputation-based
• PID Controller-based
• Multi-Krum & Krum
• Bulyan (Multi-Krum + Trimmed-Mean)
• RFA (Robust Federated Averaging)
• Trimmed-Mean

(Note: No audio — purely a visual UX showcase.)