Prototype-Guided Optimization of Quantum Data Embeddings with Linear Cost

Key points

• In supervised quantum machine learning with classical data, classification performance critically depends on the geometry induced by quantum data embedding.1
• Existing data-adaptive approaches rely on pairwise fidelity evaluations between samples, which scale as $O(B^2)$ for $B$ samples and create a computational bottleneck in iterative embedding optimization.1
• We integrate this objective into an end-to-end pipeline that combines task-specific quantum circuit architecture optimization with continuous refinement of the data-embedding parameters.1
• Across five benchmark datasets with up to 20 qubits, our method retains competitive quantum kernel support vector machine (SVM) performance while reducing fidelity-evaluation cost, with particularly strong gains over budget-matched pairwise optimization on T-MNIST.1

Sources (1)

• [1]Prototype-Guided Optimization of Quantum Data Embeddings with Linear Cost
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 17, 03:13 AM

  To address this issue, we propose a prototype-guided framework for efficient data-adaptive quantum embedding.

  In supervised quantum machine learning with classical data, classification performance critically depends on the geometry induced by quantum data embedding.

Extractive summary: sentences quoted from the sources.