FEDERATED_ARCHITECTURE.md

Federated Larun TinyML Architecture

Vision

A network of specialized TinyML nodes coordinated by a central "Grand Larun" that:

• Receives detections from edge nodes
• Aggregates and validates signals
• Shares learned patterns back to nodes
• Enables distributed exoplanet hunting

Architecture

                    ┌─────────────────────┐
                    │   GRAND LARUN       │
                    │   (Central Hub)     │
                    │   - Aggregation     │
                    │   - Validation      │
                    │   - Model Updates   │
                    └──────────┬──────────┘
                               │
           ┌───────────────────┼───────────────────┐
           │                   │                   │
    ┌──────▼──────┐     ┌──────▼──────┐     ┌──────▼──────┐
    │  TinyML-1   │     │  TinyML-2   │     │  TinyML-N   │
    │  Transit    │     │  Anomaly    │     │  Variable   │
    │  Detection  │     │  Detection  │     │  Star Class │
    └─────────────┘     └─────────────┘     └─────────────┘

Node Types

Node	TFLite Model	Purpose	Output
Transit	transit_detector.tflite	Detect dips	Period, depth, confidence
Anomaly	anomaly_detector.tflite	Flag unusual patterns	Anomaly score
Variable	variable_classifier.tflite	Classify star type	Star class
Quality	quality_scorer.tflite	Data quality check	Quality score

Communication Protocol

# Message format between nodes
{
    "node_id": "transit-001",
    "timestamp": "2026-01-31T01:00:00Z",
    "target": "TIC 307210830",
    "detection": {
        "type": "transit_candidate",
        "confidence": 0.92,
        "period": 3.5,
        "depth_ppm": 1200
    }
}

Implementation Phases

Phase 1: Specialized TFLite Models

• Train separate models for each node type
• Optimize for <100KB each

Phase 2: Node Communication

• MQTT or REST protocol
• Message queue for offline nodes

Phase 3: Grand Larun Coordinator

• Aggregation service
• Cross-validation between nodes
• Federated learning updates

Status

Approved: 2026-01-31 Priority: Future Sprint