Quantum Machine Learning — Image Classification via Quantum k-NN

This repository explores image classification using the Quantum k-Nearest Neighbors (QkNN) algorithm.
The foundation is the scheme proposed by Dang et al. (2018), which combines:

1. classical feature extraction,
2. quantum state preparation to parallelize similarity calculation,
3. quantum minimum search (Dürr–Høyer) to locate the k nearest neighbors,
4. decision through measurement.

Main reference
Image Classification Based on Quantum KNN Algorithm — Yijie Dang, Nan Jiang, Hao Hu, Zhuoxiao Ji, Wenyin Zhang (2018).
arXiv: 1805.06260 • DOI (Springer): 10.1007/s11128-018-2004-9

---

Objetivos

• Reproduzir (em simuladores) o pipeline QkNN para datasets clássicos de visão.
• Comparar complexidade e acurácia vs. k-NN clássico.
• Investigar variantes de métrica (Hamming, Euclidiana, Mahalanobis) e impacto no circuito.

Repository Structure

• notebooks/

  • 01_features_caltech101.ipynb — feature extraction from Caltech-101 dataset
  • 02_qknn_pipeline.ipynb — full quantum k-NN pipeline (state preparation, similarity, minimum search, classification)
  • 03_classical_baselines.ipynb — classical k-NN with different distance metrics
  • 99_utils_demo.ipynb — demo and utilities

• qml/

  • features.py — classical feature extraction (color histograms, texture, etc.)
  • qstates.py — quantum state preparation from feature vectors
  • distances.py — quantum distance/similarity estimation
  • minsearch.py — Dürr–Høyer quantum minimum search implementation
  • runner.py — orchestration of the QkNN pipeline

• scripts/

  • build_site.py — static site builder (converts notebooks to HTML for GitHub Pages)
  • download_data.py — dataset downloader and pre-processor

• reports/

  • evaluation results, accuracy metrics, and plots

• requirements.txt — list of dependencies

• README.md — project overview and references