Disentangling Quantum and Classical Contributions in Hybrid Quantum Machine Learning Architectures

TL;DR

The paper tackles the challenge of attributing performance in hybrid quantum-classical machine learning by proposing an autoencoder-based compression pipeline that feeds a variational quantum circuit. By training the autoencoder separately and using its encoder to supply a reduced feature set to the VQC, the authors aim to disentangle classical and quantum contributions and assess the true value of quantum processing. Empirical results across Banknote, Breast Cancer Wisconsin, MNIST, and AudioMNIST show that classical compression often dominates performance, with the quantum component offering limited or dataset-dependent gains, though angle embedding on compressed inputs matches amplitude embedding as a feasible alternative. The work highlights the need for transparent evaluation of quantum contributions in hybrid architectures and points to future directions such as joint training and broader architectural exploration to enhance quantum utility in NISQ-era tasks.

Abstract

Quantum computing offers the potential for superior computational capabilities, particularly for data-intensive tasks. However, the current state of quantum hardware puts heavy restrictions on input size. To address this, hybrid transfer learning solutions have been developed, merging pre-trained classical models, capable of handling extensive inputs, with variational quantum circuits. Yet, it remains unclear how much each component -- classical and quantum -- contributes to the model's results. We propose a novel hybrid architecture: instead of utilizing a pre-trained network for compression, we employ an autoencoder to derive a compressed version of the input data. This compressed data is then channeled through the encoder part of the autoencoder to the quantum component. We assess our model's classification capabilities against two state-of-the-art hybrid transfer learning architectures, two purely classical architectures and one quantum architecture. Their accuracy is compared across four datasets: Banknote Authentication, Breast Cancer Wisconsin, MNIST digits, and AudioMNIST. Our research suggests that classical components significantly influence classification in hybrid transfer learning, a contribution often mistakenly ascribed to the quantum element. The performance of our model aligns with that of a variational quantum circuit using amplitude embedding, positioning it as a feasible alternative.

Figures (3)

• Figure 1: Architecture of our approach, which consists of an encoder and a VQC parametrized with $\phi$. The input data is given by $\mathbb{X}$ and the prediction targets by $\hat{y}$. The VQC has $n$ qubits and consists of $\delta$ entangling layers.
• Figure 2: Validation accuracies for AudioMNIST. The classical and quantum validation accuracies are shown for the two-step hybrid transfer learning of DQC and SEQUENT.
• Figure 3: Plots for the validation accuracies for all models over the first 30 epochs. The datasets Banknote Authentication, Breast Cancer Wisconsin, MNIST, and AudioMNIST are displayed.