US Patent 10,805,353 B2

Hierarchical Federated
Security Intelligence

An open-source implementation of the patented three-tier machine learning security system that protects enterprise networks through intelligent, resource-proportional model updates.

🚀 Interactive Demo ⭐ GitHub Repository 📄 View Patent

Python 3.9+ Flask REST API 20 Patent Claims Network Security Federated ML Bank of America

System Architecture

Three-Tier Hierarchical Design

The patent describes a layered security model where updates propagate only when needed — conserving resources while maximizing security coverage.

Device Layer (Local)

Each mobile device maintains its own local security parameters. User actions update these parameters in real-time without affecting other devices.

Patent steps 305, 310, 405, 410, 435, 440

Cluster Server (Regional)

When a cluster's weighted parameter sum exceeds θ_cluster, all devices in the cluster cross-pollinate their security knowledge.

Patent steps 315–325, 415–430

Global Server (Enterprise)

When the combined cluster updates exceed θ_global, a full broadcast pushes the global model to all clusters and devices.

Patent steps 330–340, 465–470

        🌍 Global Server (params 150)

          ┣━ θ_global = 5.0

          ┣━ broadcast_to_all_clusters()

          🖥 Cluster A (params 145A)

          ┃  ┣━ θ_cluster = 1.5

          ┃  ┣━ 📱 device_1A (params 125A)

          ┃  ┗━ 📱 device_1B (params 125B)

          🖥 Cluster B (params 145B)

          ┃  ┣━ θ_cluster = 1.5

          ┃  ┣━ 📱 device_2A (params 125C)

          ┃  ┗━ 📱 device_2B (params 125D)

Key Capabilities

What FedShield Does

Every feature maps directly to a patent claim — this is a complete, faithful implementation.

⚖️

Weighted Parameter Aggregation

Assigns risk weights to each security parameter. High-risk parameters (data exfiltration, privilege escalation) carry greater weight in threshold calculations.

Claims 1, 8, 15

🔄

Cross-Cluster Knowledge Sharing

Global updates push parameters from Cluster A to Cluster B and vice versa — devices learn from threats seen by completely different network segments.

Claims 2, 3, 9, 10

⏰

Periodic Forced Updates

Even if thresholds aren't exceeded, the global server triggers mandatory updates after a configurable number of days to prevent security drift.

Claims 4, 11, 18

♻️

Failure Recovery

Device fails to update? Cluster pushes full params. Cluster fails? Global server pushes complete state. The system is self-healing by design.

Claims 5, 6, 12, 13

🚫

Real-time Action Prevention

Each device uses its current parameters to allow, flag, or block user actions in real time — without consulting a central server for every decision.

Claims 7, 14

📡

REST API Interface

Full Flask API exposing all simulation controls, state inspection, and real-time security checking. Powers the interactive demo frontend.

API endpoints

Quick Start

Get Running in 60 Seconds

# 1. Clone the repository
git clone https://github.com/prajitdatta/FedShield.git
cd FedShield

# 2. Install dependencies
pip install -r requirements.txt

# 3. Run the Python simulation directly
python src/fedshield.py

# 4. Start the REST API server
python src/api_server.py
# → API running at http://localhost:5000

# 5. Open the interactive demo (no server needed)
open demo/index.html
  

# Python usage example
from src.fedshield import FedShieldSystem

# Initialize with 2 clusters × 2 devices = 4 mobile phones
system = FedShieldSystem(
    num_clusters=2,
    devices_per_cluster=2,
    cluster_threshold=1.5,   # θ_cluster (patent step 320)
    global_threshold=5.0,    # θ_global  (patent step 335)
)

# Execute a threat simulation step
result = system.step(action_name="data_exfil", risk_delta=0.2)
print(result["update_level"])  # local | cluster | global

# Run 20 automatic steps
results = system.run_simulation(steps=20)
summary = system.get_summary()
# → { total_steps: 20, cluster_updates: 5, global_updates: 1 }
  

Patent Claims Coverage

20 Claims, All Implemented

Every claim from US Patent 10,805,353 B2 has a corresponding implementation in src/fedshield.py

1System with mobile phones, cluster servers, and global server with threshold-based parameter updates

2Cluster server propagates 3rd/4th device params back to 1st/2nd devices after global update

3Global server updates third cluster server when combined sum exceeds global threshold

4Periodic forced update: global server triggers if N days elapsed since last update

5Device failure recovery: cluster pushes all params to device that failed to update

6Cluster failure recovery: global pushes full state to cluster that failed to update

7Device uses combined params to prevent user actions that pose a security risk

8–14Method claims (same as 1–7 framed as process steps)

15–20Apparatus claims (processor + memory implementing all above)

Patent Inventor & Developer

About the Author

Prajit Kumar Datta

Patent Inventor · Named on US 10,805,353 B2 · Technology Innovator

Inventor of the hierarchical federated security learning system. This repository is the open-source codebase connected to the patent, providing a reference implementation of all 20 patent claims.