Hybrid quantum-classical circuit learning for spectral property prediction

Overview

This innovative solution leverages hybrid quantum-classical machine learning models to enhance the prediction of spectral properties. By employing parameterized quantum circuits (PQCs) for feature embedding, the approach aims to achieve improved prediction accuracy and model generalization. The use of quantum-chemical descriptors helps reduce feature space, thus optimizing the model for data-scarce scenarios. This study pioneers the application of quantum models for spectral property prediction using limited training samples, offering potential advancements in fields requiring high predictive accuracy.

Technical specifications

The research explores two main approaches:

• Encoder-Decoder Neural Network with PQCs: This method compresses quantum-chemical and steric descriptors into compact latent features optimized for quantum-enhanced processing. These features are encoded into quantum states and processed by shallow PQCs, designed to minimize NISQ-era noise. Outputs are then optimized via a classical feedback loop to predict spectral properties.

• Patch-Based Feature Encoding with PQCs: Descriptors are categorized into patches, each processed by dedicated PQCs to extract quantum-enhanced representations. These outputs are concatenated and passed to a classical neural network, maximizing qubit efficiency and enabling parallel processing.

Both approaches are implemented in Qiskit and tested through simulations and hardware, benchmarking against classical baselines on metrics like generalization error and computational overhead.

Technology readiness level

This technology is at TRL 2, indicating that the concept and application have been formulated but require further validation through experimental proof-of-concept.