Recent advancements of deep learning with the support of large-scale data sets and computational resources motivated a vast number of studies to explore deep neural networks (DNNs) in extracting information from neurophysiological data recordings such as electroencephalogram (EEG), electromyography (EMG) or electrocardiogram (ECG). Along this line, one of the major challenges in neural signal analysis is coping with the shift in data distributions across the user population and recordings, which corresponds to the widely-studied domain generalization problem for neurophysiological signals. To date, several aspects of such work focused on improving neural computer/machine interface technologies (e.g., brain-computer interfaces, robotic exoskeletons or prostheses) with state-of-the-art deep feature extraction models and domain generalization algorithms.
Adversarial machine learning so far offered a wide perspective to address domain generalization in neural signal processing. From one aspect, adversarial learning can be exploited to discover invariant neural information across data source domains. Similarly one can exploit adversarial training to learn parameterized generative distributions (such as in generative adversarial networks) for neurophysiological recording data sets, where artificial samples could be synthesized to close the domain generalization gap in calibration. On another line of work, adversarial machine learning also considers security vulnerabilities of DNN-driven neural interface technologies, where inference pipelines are manipulated via so-called adversarial attacks (i.e., lack of generalization to minimally-perturbed adversarial examples). While substantial progress has been made in various aspects, there are several open problems remaining to be addressed. The goal of this Research Topic is to present latest advances in adversarial learning and domain generalization methods tailored to various aspects of neural signal analysis.
In this Research Topic we aim to solicit high-quality papers that report emerging developments at the intersection of adversarial machine learning and domain generalization, with applications to neurophysiological signal analysis. Potential authors are invited to submit original research contributions, as well as review papers. Topics of interest may include, but are not limited to:
- Adversarial machine learning for neural signal processing
- Novel deep time-series signal analysis architectures and optimization strategies
- Deep generative modeling for neural data augmentation
- Representation learning methods and architectures for time-series neural recordings
- User- or domain-invariant representation learning from neurophysiological signals
- Cross-domain invariance and generalization methods (e.g., users, recording sessions)
- Adversarial attacks and defenses to neural signal classification models
- Adversarial training for enhanced robustness in neural signal analysis
- Graph neural networks for domain generalization in neural signal analysis
- Self-supervised learning methods for domain generalization with neural signals
- Explainable deep learning models for neurophysiological data
- User-independent calibration and generalization for online applications
- Interdisciplinary applications (e.g., robotic neural interfaces)
Recent advancements of deep learning with the support of large-scale data sets and computational resources motivated a vast number of studies to explore deep neural networks (DNNs) in extracting information from neurophysiological data recordings such as electroencephalogram (EEG), electromyography (EMG) or electrocardiogram (ECG). Along this line, one of the major challenges in neural signal analysis is coping with the shift in data distributions across the user population and recordings, which corresponds to the widely-studied domain generalization problem for neurophysiological signals. To date, several aspects of such work focused on improving neural computer/machine interface technologies (e.g., brain-computer interfaces, robotic exoskeletons or prostheses) with state-of-the-art deep feature extraction models and domain generalization algorithms.
Adversarial machine learning so far offered a wide perspective to address domain generalization in neural signal processing. From one aspect, adversarial learning can be exploited to discover invariant neural information across data source domains. Similarly one can exploit adversarial training to learn parameterized generative distributions (such as in generative adversarial networks) for neurophysiological recording data sets, where artificial samples could be synthesized to close the domain generalization gap in calibration. On another line of work, adversarial machine learning also considers security vulnerabilities of DNN-driven neural interface technologies, where inference pipelines are manipulated via so-called adversarial attacks (i.e., lack of generalization to minimally-perturbed adversarial examples). While substantial progress has been made in various aspects, there are several open problems remaining to be addressed. The goal of this Research Topic is to present latest advances in adversarial learning and domain generalization methods tailored to various aspects of neural signal analysis.
In this Research Topic we aim to solicit high-quality papers that report emerging developments at the intersection of adversarial machine learning and domain generalization, with applications to neurophysiological signal analysis. Potential authors are invited to submit original research contributions, as well as review papers. Topics of interest may include, but are not limited to:
- Adversarial machine learning for neural signal processing
- Novel deep time-series signal analysis architectures and optimization strategies
- Deep generative modeling for neural data augmentation
- Representation learning methods and architectures for time-series neural recordings
- User- or domain-invariant representation learning from neurophysiological signals
- Cross-domain invariance and generalization methods (e.g., users, recording sessions)
- Adversarial attacks and defenses to neural signal classification models
- Adversarial training for enhanced robustness in neural signal analysis
- Graph neural networks for domain generalization in neural signal analysis
- Self-supervised learning methods for domain generalization with neural signals
- Explainable deep learning models for neurophysiological data
- User-independent calibration and generalization for online applications
- Interdisciplinary applications (e.g., robotic neural interfaces)