Modeling Feature Maps for Quantum Machine Learning

TL;DR

The paper tackles the challenge of quantum noise on near-term quantum devices for genomic sequence classification, proposing a systematic evaluation across multiple noise models and three quantum feature maps. By modeling four QML algorithms—QSVC, Peg-QSVC, QNN, and VQC—under dephasing, amplitude damping, depolarizing, thermal relaxation, bit-flip, and phase-flip noise, it reveals that QSVC generally exhibits the strongest robustness while Peg-QSVC and QNN are more sensitive to noise, particularly with entangled feature maps. The study highlights the critical role of feature-map selection in noisy QML, showing a trade-off between expressivity and noise resilience: ZFeatureMap is robust but less expressive, whereas ZZFeatureMap and PauliFeatureMap offer richer representations yet are more vulnerable to noise, especially under depolarizing and amplitude-damping channels. These insights provide practical guidance for noise-aware design of QML pipelines in genomics and point to the need for targeted noise-mitigation strategies to enable reliable quantum genome analyses on NISQ hardware.

Abstract

Quantum Machine Learning (QML) offers significant potential for complex tasks like genome sequence classification, but quantum noise on Noisy Intermediate-Scale Quantum (NISQ) devices poses practical challenges. This study systematically evaluates how various quantum noise models including dephasing, amplitude damping, depolarizing, thermal noise, bit-flip, and phase-flip affect key QML algorithms (QSVC, Peg-QSVC, QNN, VQC) and feature mapping techniques (ZFeatureMap, ZZFeatureMap, and PauliFeatureMap). Results indicate that QSVC is notably robust under noise, whereas Peg-QSVC and QNN are more sensitive, particularly to depolarizing and amplitude-damping noise. The PauliFeatureMap is especially vulnerable, highlighting difficulties in maintaining accurate classification under noisy conditions. These findings underscore the critical importance of feature map selection and noise mitigation strategies in optimizing QML for genomic classification, with promising implications for personalized medicine.

Figures (14)

• Figure 1: Overview of the proposed workflow presented in this paper. a) Dataset Split: Split the classical dataset into training and testing subsets. b) Classical Dimensionality Reduction: Apply PCA to reduce the dataset to four dimensions. c) Feature Mapping: Transform the dataset into quantum states into in Hilbert space. The performance of these encoding techniques is notably influenced by different types of inherent quantum noise in NISQ devices. d) QML Algorithm: Trains various QML algorithms on the encoded quantum data, with quantum noise affecting the training process and algorithm performance. e) Evaluation: Assess the impact of quantum noise on encoding by generating quantum states and using the trained QML models to classify test sequences.
• Figure 2: Output state under no noise conditions.
• Figure 3: Output state under depolarizing noise.
• Figure 4: Output state under amplitude damping noise.
• Figure 5: Output state under phase damping noise.