Deep Source Semi-Supervised Transfer Learning (DS3TL) for Cross-Subject EEG Classification.

OBJECTIVE: An electroencephalogram (EEG) based brain-computer interface (BCI) maps the user's EEG signals into commands for external device control. Usually a large amount of labeled EEG trials are required to train a reliable EEG recognition model. However, acquiring labeled EEG data is time-consuming and user-unfriendly. Semi-supervised learning (SSL) and transfer learning can be used to exploit the unlabeled data and the auxiliary data, respectively, to reduce the amount of labeled data for a new subject. METHODS: This paper proposes deep source semi-supervised transfer learning (DS3TL) for EEG-based BCIs, which assumes the source subject has a small number of labeled EEG trials and a large number of unlabeled ones, whereas all EEG trials from the target subject are unlabeled. DS3TL mainly includes a hybrid SSL module, a weakly-supervised contrastive module, and a domain adaptation module. The hybrid SSL module integrates pseudo-labeling and consistency regularization for SSL. The weakly-supervised contrastive module performs contrastive learning by using the true labels of the labeled data and the pseudo-labels of the unlabeled data. The domain adaptation module reduces the individual differences by uncertainty reduction. RESULTS: Experiments on three EEG datasets from different tasks demonstrated that DS3TL outperformed a supervised learning baseline with many more labeled training data, and multiple state-of-the-art SSL approaches with the same number of labeled data. SIGNIFICANCE: To our knowledge, this is the first approach in EEG-based BCIs that exploits the unlabeled source data for more accurate target classifier training.

User Identity Protection in EEG-Based Brain-Computer Interfaces.

A brain-computer interface (BCI) establishes a direct communication pathway between the brain and an external device. Electroencephalogram (EEG) is the most popular input signal in BCIs, due to its convenience and low cost. Most research on EEG-based BCIs focuses on the accurate decoding of EEG signals; however, EEG signals also contain rich private information, e.g., user identity, emotion, and so on, which should be protected. This paper first exposes a serious privacy problem in EEG-based BCIs, i.e., the user identity in EEG data can be easily learned so that different sessions of EEG data from the same user can be associated together to more reliably mine private information. To address this issue, we further propose two approaches to convert the original EEG data into identity-unlearnable EEG data, i.e., removing the user identity information while maintaining the good performance on the primary BCI task. Experiments on seven EEG datasets from five different BCI paradigms showed that on average the generated identity-unlearnable EEG data can reduce the user identification accuracy from 70.01% to at most 21.36%, greatly facilitating user privacy protection in EEG-based BCIs.

EEG-Based Brain-Computer Interfaces are Vulnerable to Backdoor Attacks.

Research and development of electroencephalogram (EEG) based brain-computer interfaces (BCIs) have advanced rapidly, partly due to deeper understanding of the brain and wide adoption of sophisticated machine learning approaches for decoding the EEG signals. However, recent studies have shown that machine learning algorithms are vulnerable to adversarial attacks. This paper proposes to use narrow period pulse for poisoning attack of EEG-based BCIs, which makes adversarial attacks much easier to implement. One can create dangerous backdoors in the machine learning model by injecting poisoning samples into the training set. Test samples with the backdoor key will then be classified into the target class specified by the attacker. What most distinguishes our approach from previous ones is that the backdoor key does not need to be synchronized with the EEG trials, making it very easy to implement. The effectiveness and robustness of the backdoor attack approach is demonstrated, highlighting a critical security concern for EEG-based BCIs and calling for urgent attention to address it.

Universal adversarial perturbations for CNN classifiers in EEG-based BCIs.

Objective. Multiple convolutional neural network (CNN) classifiers have been proposed for electroencephalogram (EEG) based brain-computer interfaces (BCIs). However, CNN models have been found vulnerable to universal adversarial perturbations (UAPs), which are small and example-independent, yet powerful enough to degrade the performance of a CNN model, when added to a benign example.Approach. This paper proposes a novel total loss minimization (TLM) approach to generate UAPs for EEG-based BCIs.Main results. Experimental results demonstrated the effectiveness of TLM on three popular CNN classifiers for both target and non-target attacks. We also verified the transferability of UAPs in EEG-based BCI systems.Significance. To our knowledge, this is the first study on UAPs of CNN classifiers in EEG-based BCIs. UAPs are easy to construct, and can attack BCIs in real-time, exposing a potentially critical security concern of BCIs.