| name | tensor-network-quantum-federated |
| description | Privacy-aware federated learning combining tensor-network compression with quantum-enhanced processing for medical diagnosis. Use when building multi-institutional medical AI systems that need: (1) MPC-secured aggregation, (2) small-qubit quantum processing on compressed features, (3) tensor-network frontends (MPS/TTN/MERA). Addresses the dual challenge of communication overhead and qubit limitations.
|
Tensor-Network Quantum Federated Learning
Architecture Overview
[Client A] ── MPS/TTN/MERA ──┐
[Client B] ── MPS/TTN/MERA ──┤ → MPC Aggregation → QEP → Diagnosis
[Client C] ── MPS/TTN/MERA ──┘
Three-Layer Design
Layer 1: Tensor-Network Frontend (Client-Side)
Compresses high-dimensional medical images into compact latent representations.
| Frontend | Compression | Best For |
|---|
| MPS | Linear scaling | 1D sequences, time-series |
| TTN | Logarithmic scaling | 2D images (recommended) |
| MERA | Multi-scale | Hierarchical features |
Key insight: TTN+QEP combination shows the most balanced overall profile.
Layer 2: MPC-Secured Aggregation (Server-Side)
- Multi-party computation protects aggregated latents
- Communication cost ∝ latent dimension (not original image size)
- Tensor-network compression directly reduces MPC overhead
Layer 3: Quantum-Enhanced Processor (Post-Aggregation)
- Quantum-state embedding of aggregated latents
- Observable-based readout for classification
- Stable when qubit count ≈ latent dimension
- Degrades under noise vs. noiseless simulation
Design Principles
