Exploring the Potential of QEEGNet for Cross-Task and Cross-Dataset Electroencephalography Encoding with Quantum Machine Learning

TL;DR

The paper tackles EEG decoding in the presence of inter-subject variability and noise by introducing QEEGNet, a hybrid quantum-classical extension of EEGNet that incorporates a quantum encoding layer and a variational quantum circuit. Using four public EEG datasets (motor imagery and P300 paradigms), the authors evaluate cross-task and cross-dataset generalization, showing that QEEGNet can achieve competitive or improved recall and F1 scores in several cases, though gains are not universal across all datasets. The architecture adds a linear quantum overhead, maintaining computational efficiency similar to EEGNet, which is important for practical feasibility. The study provides valuable insights into the opportunities and challenges of applying quantum machine learning to EEG analysis and outlines directions for circuit design and large-scale validation.

Abstract

Electroencephalography (EEG) is widely used in neuroscience and clinical research for analyzing brain activity. While deep learning models such as EEGNet have shown success in decoding EEG signals, they often struggle with data complexity, inter-subject variability, and noise robustness. Recent advancements in quantum machine learning (QML) offer new opportunities to enhance EEG analysis by leveraging quantum computing's unique properties. In this study, we extend the previously proposed Quantum-EEGNet (QEEGNet), a hybrid neural network incorporating quantum layers into EEGNet, to investigate its generalization ability across multiple EEG datasets. Our evaluation spans a diverse set of cognitive and motor task datasets, assessing QEEGNet's performance in different learning scenarios. Experimental results reveal that while QEEGNet demonstrates competitive performance and maintains robustness in certain datasets, its improvements over traditional deep learning methods remain inconsistent. These findings suggest that hybrid quantum-classical architectures require further optimization to fully leverage quantum advantages in EEG processing. Despite these limitations, our study provides new insights into the applicability of QML in EEG research and highlights challenges that must be addressed for future advancements.

Figures (5)

• Figure 1: Hybrid Quantum-Classical Framework for EEG Processing in QEEGNet. The framework integrates a classical neural network with a variational quantum circuit to process electroencephalography (EEG) data. EEG features are first extracted and transformed by a classical neural network, parameterized by $W$. The resulting feature representations are then encoded into a quantum state and processed using a quantum neural network with trainable parameters $\Theta$. The quantum measurement outcomes are fed back into a classical computer, updating both quantum and classical parameters iteratively. This hybrid approach leverages quantum computing to enhance feature representation learning for EEG-based applications.
• Figure 2: Structure of the Variational Quantum Circuit (VQC). The input $\vec{x}$ represents EEG features encoded by a classical neural network before being processed by the quantum circuit. The circuit consists of an encoding layer $U(C(\vec{x}))$, variational layers $V_i(\vec{\theta}_i)$, and measurement operations to extract quantum outputs.
• Figure 3: The VQC (QNN) layer in this work.
• Figure 4: A schematic comparison between EEGNet and the proposed Quantum-EEGNet (QEEGNet). In this illustration, Conv represents convolution operations, BN stands for batch normalization, and ELU refers to the exponential linear unit activation function.
• Figure 5: Comparison of feature space visualization of EEGNet and QEEGNet for Subject 9 in the PhysioP300 dataset using t-SNE and UMAP. (a) EEGNet visualization, showing the learned feature separability between target (blue) and non-target (red) samples. (b) QEEGNet visualization, demonstrating improved clustering of target and non-target samples compared to EEGNet. The clearer separation in QEEGNet’s feature space suggests its enhanced ability to distinguish P300 responses, highlighting its potential for more robust classification.