quantum-prototype-learning

name	quantum-prototype-learning
description	Geometric prototype learning in quantum Hilbert space using matrix product states for explainable ML
category	ai_collection

Quantum Prototype Learning in Hilbert Space

Description

Geometric prototype learning methodology for quantum machine learning in Hilbert space. Class representatives are encoded as generative matrix product states (MPS) residing in the same Hilbert space as quantum-encoded data samples. Classification and clustering are performed through geometric measures of quantum states. Lifts prototype learning from classical feature space to quantum Hilbert space, providing an explainable framework for ML.

Activation Keywords

• quantum prototype learning
• geometric prototype
• matrix product state ML
• Hilbert space learning
• quantum state classification
• 量子原型学习
• 希尔伯特空间学习
• MPS classification
• quantum geometric measures

Core Concepts

Matrix Product State (MPS) Prototypes

• Each class is represented by a generative MPS
• MPS serves as a compressed, structured representation of class data
• Prototypes are interpretable — unlike black-box neural network weights

Quantum Hilbert Space Learning

• Data and prototypes reside in the same Hilbert space
• Classification via geometric distance/overlap between quantum states
• Inner products, fidelity, trace distance as similarity measures

Attraction Effect

• Quantum-probabilistic prototypes induce an "attraction" effect
• Points near prototypes are pulled closer in the quantum metric
• Provides natural clustering behavior without explicit cluster assignments

Dimensionality Reduction via Prototype Distances

• Distances to prototypes serve as low-dimensional features
• More interpretable than PCA/autoencoder embeddings
• Preserves class-relevant geometric structure

Usage Patterns

Pattern 1: Quantum Prototype Classification

1. Encode training data as quantum states (tensor network representation)
2. Learn one MPS prototype per class via optimization
3. Classify new samples by geometric proximity to prototypes
4. Output: class label + distance to each prototype (explainable)

Pattern 2: Quantum Prototype Clustering

1. Initialize k MPS prototypes
2. Assign each data point to nearest prototype (quantum metric)
3. Update prototypes to minimize intra-cluster quantum distance
4. Iterate until convergence

Pattern 3: Dimensionality Reduction

1. Train k class prototypes as MPS
2. Map each sample to vector of k prototype distances
3. Use reduced representation for downstream tasks

Implementation Guidelines

MPS Bond Dimension Selection

• Start with small bond dimension (e.g., D=4-8)
• Increase until validation accuracy plateaus
• Larger D → more expressive but harder to train

Geometric Measures

• Fidelity: F(ρ₁, ρ₂) = Tr√(√ρ₁ ρ₂ √ρ₁) for mixed states
• Overlap: |⟨ψ₁|ψ₂⟩|² for pure states
• Trace distance: D(ρ₁, ρ₂) = ½‖ρ₁ - ρ₂‖₁

Training Objective

• Minimize distance between same-class samples and their prototype
• Maximize distance between different-class samples and wrong prototypes
• Regularize bond dimension for compression

Error Handling

Overfitting with Large Bond Dimension

• Use cross-validation to select optimal bond dimension
• Apply early stopping based on validation performance
• Consider tensor network regularization

Training Instability

• Initialize prototypes as class-mean MPS
• Use Riemannian optimization on MPS manifold
• Normalize MPS at each step for numerical stability

Computational Complexity

• MPS operations scale as O(N·D²) where N=system size, D=bond dimension
• For large datasets, use batched MPS updates
• Consider truncated SVD for efficient MPS operations

Benchmarks

• Fashion-MNIST: Outperforms classical prototype methods
• ECG dataset: Competitive with black-box neural networks
• Provides explainability advantage over standard classifiers

References

• arXiv:2605.17895 - Geometric Prototype Learning in Quantum Hilbert Space with Matrix Product States
• Tensor network machine learning
• Matrix product state literature

arXiv Reference

• Paper: Geometric Prototype Learning in Quantum Hilbert Space with Matrix Product States
• ID: 2605.17895
• Date: 2026-05-18
• Authors: Kun Zhang, Lei Ding, Sheng-Chen Bai, Jing Sun, An-Qi Jing, Min Tang, Shi-Ju Ran