Home▸Machine Learning & Neural Networks▸Byzantine-Robust Federated Aggregation

Byzantine-Robust Federated Aggregation

Interactive 3D simulator of Byzantine-robust federated learning: watch malicious clients send poisoned model updates, then compare how plain FedAvg, coordinate-wise median, trimmed mean and Krum aggregation each resist or fall for the attack.

Machine Learning & Neural Networks3DAdvanced60 FPS📱 Mobile-adapted

Details

CategoryMachine Learning & Neural Networks

RenderingThree.js · WebGL

DifficultyAdvanced

Target FPS60

ControlsButtons switch the aggregation rule (FedAvg / Median / Trimmed Mean / Krum); sliders set total client count, number of malicious clients and attack strength; Run Round advances one federated round, Auto-run loops rounds continuously, and Reset Training restarts the global model from its starting position.

federated-learningbyzantine-fault-tolerancerobust-statisticsmachine-learningkrumdistributed-systems

Related simulations

Secure Aggregation in Federated LearningInteractive 3D simulator of the pairwise-masking secure-aggregation protocol used in…

Federated Learning Communication Efficiency Simulator (2D)Watch 10 federated-learning clients compress their model updates with top-K sparsification…

Secure Aggregation in Federated LearningInteractive 3D simulator of the pairwise-masking secure-aggregation protocol used in… ⊞ All 226 in Machine Learning & Neural Networks →

Federated learning trains one shared model from many clients' local updates without ever collecting raw data — but that also means the server must trust update vectors it cannot inspect. This simulator shows what happens when some clients are malicious: it renders every client's update as an arrow in a 3D parameter space, lets you dial in how many clients are compromised and how aggressively they attack, and compares four real aggregation rules — plain FedAvg, coordinate-wise median, trimmed mean and Krum — to see which ones keep the global model converging toward the true optimum and which get dragged off course by a handful of poisoned updates.

⚙ Under the hood

Interactive 3D simulator of Byzantine-robust federated learning: malicious clients send poisoned updates while you compare how plain FedAvg, coordinate-wise median, trimmed mean and Krum aggregation each resist or fall for the attack.

federated-learningbyzantine-fault-tolerancerobust-statisticsmachine-learningkrumdistributed-systems

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install