skills.rest
maruakshay
maruakshayCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

federated-learning-security

Detect poisoned updates, tampering, and gradient leakage in federated learning systems.

4|Updated Apr 27, 2026
Search Tags:#privacy-preserving#server-hardening#federated-learning#secure-aggregation#byzantine-robustness#gradient-poisoning
One-click install
npx skills add https://github.com/maruakshay/mii-ai-security --skill federated-learning-security
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: federated-learning-security
Source: https://github.com/maruakshay/mii-ai-security/tree/main/skills/federated-learning-security
Command: npx skills add https://github.com/maruakshay/mii-ai-security --skill federated-learning-security

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Federated learning security reviews mitigate poisoned gradient updates, model tampering by malicious participants, aggregation server compromises, Byzantine fault tolerance gaps, and privacy leakage via gradient inversion.

Core Features & Use Cases

  • Validate and harden aggregation against Byzantine faults using robust aggregators (e.g., coordinate-wise median, Krum) and prevent naive FedAvg domination.
  • Enforce privacy protections with secure aggregation, local differential privacy, and hardened server controls (mTLS, audit logs, signed checkpoints).
  • Apply per-participant anomaly detection and monitoring across rounds to detect consistently dissimilar updates.

Quick Start

Review a federated learning deployment and implement robust aggregation, privacy protections, and server hardening in your ML workflow.

Frequently Asked Questions about federated-learning-security

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I detect poisoned gradient updates in federated learning?▼

Detect poisoned gradient updates in federated learning by applying per-participant anomaly tracking across rounds and Byzantine-robust aggregators like coordinate-wise median or Krum to isolate consistently dissimilar and malicious model updates.

What is Byzantine-robust aggregation and how does it prevent model tampering?▼

Byzantine-robust aggregation prevents model tampering by replacing naive FedAvg with robust aggregators like coordinate-wise median or Krum, ensuring that malicious participant updates cannot dominate the global model during federated learning rounds.

How do I prevent gradient leakage and protect privacy in federated learning?▼

Prevent gradient leakage and protect privacy in federated learning by enforcing secure aggregation protocols, applying local differential privacy, and implementing gradient norm clipping to mitigate gradient inversion attacks.

What server hardening measures are required for a secure federated learning deployment?▼

Server hardening measures for a secure federated learning deployment include enforcing mutual TLS (mTLS) for participant communication, maintaining comprehensive audit logs, and cryptographically signing global model checkpoints to prevent tampering.

Does this federated learning security review support cross-silo and cross-device environments?▼

Yes, this federated learning security review is explicitly applicable to both cross-device and cross-silo environments, validating robust aggregation, privacy guarantees, and server hardening across diverse ML workloads and deployment scales.

Why should I replace naive FedAvg with robust aggregators in federated learning?▼

Replace naive FedAvg with robust aggregators because standard averaging is highly vulnerable to Byzantine faults and poisoned updates, allowing a single malicious participant to manipulate the global model during federated training.

Related Skills

Explore More Software Engineering Skills→
hiyenwonghiyenwong
2

heterogeneous-stochastic-momentum-admm

Solve distributed non-convex composite optimization on heterogeneous networks with adaptive momentum ADMM.

Community
Advanced
shafibabarshafibabar
1

compliance-verification

Automate security and compliance verification with signed attestations and control gates.

Community
Advanced
mitchuskimitchuski

agentprivacy-intel-pooling

Aggregate agent observations into collective intelligence without source disclosure.

Community
Advanced
NVIDIANVIDIA
2.8k

resiliency

Integrate nvidia-resiliency-ext fault tolerance into Megatron Bridge training workflows.

Official
Advanced