FLAVOR

Federated Learning Analytics, Visualization, Optimization & Reliability

FLAVOR is a tool for visualizing and understanding the privacy-preserving properties and optimization techniques used in federated learning.

Choose Your Experience

Select a learning path based on your expertise and goals

Beginner Friendly

Simple Workflow

Educational Demonstration

A guided, pedagogical walkthrough of federated learning concepts with static visualizations. Perfect for learning the fundamentals.

Features:

✓Step-by-step educational flow
✓Visual explanations of FL concepts
✓Shamir Secret Sharing demo
✓FedAvg aggregation basics

5-7 minutes

Advanced

Bonawitz Protocol (2017)

Interactive Secure Aggregation

Bonawitz et al., ACM CCS 2017 (arXiv:1611.04482)

Full implementation of the Bonawitz et al. secure aggregation protocol with interactive client simulation and dropout recovery.

Features:

✓Interactive protocol simulation
✓Pairwise Diffie-Hellman key exchange
✓Dropout-resistant aggregation
✓Shamir Secret Sharing for recovery
✓Real-time mask generation & cancellation

7-10 minutes

Intermediate

Gradient Inversion Attack (2019)

How Shared Gradients Can Reveal Private Training Data

Zhu et al., NeurIPS 2019 (arXiv:1906.08935)

Interactive demonstration of the Gradient Inversion (DLG) attack (Zhu et al., NeurIPS 2019) showing how shared gradients can leak private training data.

Features:

✓Live gradient inversion attack demo
✓Multiple CNN architectures (Simple/LeNet/ResNet)
✓Batch size experiments (1, 2, 4, 8)
✓Interactive defense mechanisms
✓Secure aggregation as defense

7-10 minutes

More Protocols Coming Soon

Future implementations will include:

SecAgg+ (2020)LightSecAgg (2022)FastSecAgg (2023)Custom Protocols

Built for education and research • Interactive FL protocol demonstrations

Made by Aurelius Nguyen with ❤️