In the field of brain-computer interfaces (BCIs), the potential for leveraging deep learning techniques for representing electroencephalogram (EEG) signals has gained substantial interest. This review synthesizes empirical findings from a collection of articles using deep representation learning techniques for BCI decoding, to provide a comprehensive analysis of the current state-of-the-art. Each article was scrutinized based on three criteria: (1) the deep representation learning technique employed, (2) the underlying motivation for its utilization, and (3) the approaches adopted for characterizing the learned representations. Among the 81 articles finally reviewed in depth, our analysis reveals a predominance of 31 articles using autoencoders. We identified 13 studies employing self-supervised learning (SSL) techniques, among which ten were published in 2022 or later, attesting to the relative youth of the field. However, at the time being, none of these have led to standard foundation models that are picked up by the BCI community. Likewise, only a few studies have introspected their learned representations. We observed that the motivation in most studies for using representation learning techniques is for solving transfer learning tasks, but we also found more specific motivations such as to learn robustness or invariances, as an algorithmic bridge, or finally to uncover the structure of the data. Given the potential of foundation models to effectively tackle these challenges, we advocate for a continued dedication to the advancement of foundation models specifically designed for EEG signal decoding by using SSL techniques. We also underline the imperative of establishing specialized benchmarks and datasets to facilitate the development and continuous improvement of such foundation models.

In the BCI field, introspection and interpretation of brain signals are desired for providing feedback or to guide rapid paradigm prototyping but are challenging due to the high noise level and dimensionality of the signals. Deep neural networks are often introspected by transforming their learned feature representations into 2- or 3-dimensional subspace visualizations using projection algorithms like Uniform Manifold Approximation and Projection (UMAP). Unfortunately, these methods are computationally expensive, making the projection of data streams in real-time a non-trivial task. In this study, we introduce a novel variant of UMAP, called approximate UMAP (aUMAP). It aims at generating rapid projections for real-time introspection. To study its suitability for real-time projecting, we benchmark the methods against standard UMAP and its neural network counterpart parametric UMAP. Our results show that approximate UMAP delivers projections that replicate the projection space of standard UMAP while decreasing projection speed by an order of magnitude and maintaining the same training time.

Data scarcity in the brain-computer interface field can be alleviated through the use of generative models, specifically diffusion models. While diffusion models have previously been successfully applied to electroencephalogram (EEG) data, existing models lack flexibility w.r.t.~sampling or require alternative representations of the EEG data. To overcome these limitations, we introduce a novel approach to conditional diffusion models that utilizes classifier-free guidance to directly generate subject-, session-, and class-specific EEG data. In addition to commonly used metrics, domain-specific metrics are employed to evaluate the specificity of the generated samples. The results indicate that the proposed model can generate EEG data that resembles real data for each subject, session, and class.

Motivated by the challenge of seamless cross-dataset transfer in EEG signal processing, this article presents an exploratory study on the use of Joint Embedding Predictive Architectures (JEPAs). In recent years, self-supervised learning has emerged as a promising approach for transfer learning in various domains. However, its application to EEG signals remains largely unexplored. In this article, we introduce Signal-JEPA for representing EEG recordings which includes a novel domain-specific spatial block masking strategy and three novel architectures for downstream classification. The study is conducted on a 54~subjects dataset and the downstream performance of the models is evaluated on three different BCI paradigms: motor imagery, ERP and SSVEP. Our study provides preliminary evidence for the potential of JEPAs in EEG signal encoding. Notably, our results highlight the importance of spatial filtering for accurate downstream classification and reveal an influence of the length of the pre-training examples but not of the mask size on the downstream performance.

We present a simple deep learning-based framework commonly used in computer vision and demonstrate its effectiveness for cross-dataset transfer learning in mental imagery decoding tasks that are common in the field of Brain-Computer Interfaces (BCI). We investigate, on a large selection of 12 motor-imagery datasets, which ones are well suited for transfer, both as donors and as receivers. Challenges. Deep learning models typically require long training times and are data-hungry, which impedes their use for BCI systems that have to minimize the recording time for (training) examples and are subject to constraints induced by experiments involving human subjects. A solution to both issues is transfer learning, but it comes with its own challenge, i.e., substantial data distribution shifts between datasets, subjects and even between subsequent sessions of the same subject. Approach. For every pair of pre-training (donor) and test (receiver) dataset, we first train a model on the donor before training merely an additional new linear classification layer based on a few receiver trials. Performance of this transfer approach is then tested on other trials of the receiver dataset. Significance. First, we lower the threshold to use transfer learning between motor imagery datasets: the overall framework is extremely simple and nevertheless obtains decent classification scores. Second, we demonstrate that deep learning models are a good option for motor imagery cross-dataset transfer both for the reasons outlined in the first point and because the framework presented is viable in online scenarios. Finally, analysing which datasets are best suited for transfer learning can be used as a reference for future researchers to determine which to use for pre-training or benchmarking.

Many brain-computer interfaces make use of brain signals that are elicited in response to a visual, auditory or tactile stimulus, so-called event-related potentials (ERPs). In visual ERP speller applications, sets of letters shown on a screen are flashed randomly, and the participant attends to the target letter they want to spell. When this letter flashes, the resulting ERP is different compared to when any other non-target letter flashes. We propose a new unsupervised approach to detect this attended letter. In each trial, for every available letter our approach makes the hypothesis that it is in fact the attended letter, and calculates the ERPs based on each of these hypotheses. We leverage the fact that only the true hypothesis produces the largest difference between the class means. Note that this unsupervised method does not require any changes to the underlying experimental paradigm and therefore can be employed in almost any ERP-based setup. To deal with limited data, we use a block-Toeplitz regularized covariance matrix that models the background activity. We implemented the proposed novel unsupervised mean-difference maximization (UMM) method and evaluated it in offline replays of brain-computer interface visual speller datasets. For a dataset that used 16 flashes per symbol per trial, UMM correctly classifies 3651 out of 3654 letters ($99.92\,\%$) across 25 participants. In another dataset with fewer and shorter trials, 7344 out of 7383 letters ($99.47\,\%$) are classified correctly across 54 participants with two sessions each. Even in more challenging datasets obtained from patients with amyotrophic lateral sclerosis ($77.86\,\%$) or when using auditory ERPs ($82.52\,\%$), the obtained classification rates obtained by UMM are competitive. In addition, UMM provides stable confidence measures which can be used to monitor convergence.

Neurophysiological time series recordings like the electroencephalogram (EEG) or local field potentials are obtained from multiple sensors. They can be decoded by machine learning models in order to estimate the ongoing brain state of a patient or healthy user. In a brain-computer interface (BCI), this decoded brain state information can be used with minimal time delay to either control an application, e.g., for communication or for rehabilitation after stroke, or to passively monitor the ongoing brain state of the subject, e.g., in a demanding work environment. A specific challenge in such decoding tasks is posed by the small dataset sizes in BCI compared to other domains of machine learning like computer vision or natural language processing. A possibility to tackle classification or regression problems in BCI despite small training data sets is through transfer learning, which utilizes data from other sessions, subjects or even datasets to train a model. In this exploratory study, we propose novel domain-specific embeddings for neurophysiological data. Our approach is based on metric learning and builds upon the recently proposed ladder loss. Using embeddings allowed us to benefit, both from the good generalisation abilities and robustness of deep learning and from the fast training of classical machine learning models for subject-specific calibration. In offline analyses using EEG data of 14 subjects, we tested the embeddings' feasibility and compared their efficiency with state-of-the-art deep learning models and conventional machine learning pipelines. In summary, we propose the use of metric learning to obtain pre-trained embeddings of EEG-BCI data as a means to incorporate domain knowledge and to reach competitive performance on novel subjects with minimal calibration requirements.

Robot motions in the presence of humans should not only be feasible and safe, but also conform to human preferences. This, however, requires user feedback on the robot's behavior. In this work, we propose a novel approach to leverage the user's brain signals as a feedback modality in order to decode the judgment of robot trajectories and rank them according to the user's preferences. We show that brain signals measured using electroencephalography during observation of a robotic arm's trajectory as well as in response to preference statements are informative regarding the user's preference. Furthermore, we demonstrate that user feedback from brain signals can be used to reliably infer pairwise trajectory preferences as well as to retrieve the preferred observed trajectories of the user with a performance comparable to explicit behavioral feedback.

Data-driven spatial filtering approaches are commonly used to assess rhythmic brain activity from multichannel recordings such as electroencephalography (EEG). As spatial filter estimation is prone to noise, non-stationarity effects and limited data, a high model variability induced by slight changes of, e.g., involved hyperparameters is generally encountered. These aspects challenge the assessment of functionally relevant features which are of special importance in closed-loop applications as, e.g., in the field of rehabilitation. We propose a data-driven method to identify groups of reliable and functionally relevant oscillatory components computed by a spatial filtering approach. Therefore, we initially embrace the variability of decoding models in a large configuration space before condensing information by density-based clustering of components' functional signatures. Exemplified for a hand force task with rich within-trial structure, the approach was evaluated on EEG data of 18 healthy subjects. We found that functional characteristics of single components are revealed by distinct temporal dynamics of their event-related power changes. Based on a within-subject analysis, our clustering revealed seven groups of homogeneous envelope dynamics on average. To support introspection by practitioners, we provide a set of metrics to characterize and validate single clusterings. We show that identified clusters contain components of strictly confined frequency ranges, dominated by the alpha and beta band. Our method is applicable to any spatial filtering algorithm. Despite high model variability, it allows capturing and monitoring relevant oscillatory features. We foresee its application in closed-loop applications such as brain-computer interface based protocols in stroke rehabilitation.

Many cognitive, sensory and motor processes have correlates in oscillatory neural sources, which are embedded as a subspace into the recorded brain signals. Decoding such processes from noisy magnetoencephalogram/electroencephalogram (M/EEG) signals usually requires the use of data-driven analysis methods. The objective evaluation of such decoding algorithms on experimental raw signals, however, is a challenge: the amount of available M/EEG data typically is limited, labels can be unreliable, and raw signals often are contaminated with artifacts. The latter is specifically problematic, if the artifacts stem from behavioral confounds of the oscillatory neural processes of interest. To overcome some of these problems, simulation frameworks have been introduced for benchmarking decoding methods. Generating artificial brain signals, however, most simulation frameworks make strong and partially unrealistic assumptions about brain activity, which limits the generalization of obtained results to real-world conditions. In the present contribution, we thrive to remove many shortcomings of current simulation frameworks and propose a versatile alternative, that allows for objective evaluation and benchmarking of novel data-driven decoding methods for neural signals. Its central idea is to utilize post-hoc labelings of arbitrary M/EEG recordings. This strategy makes it paradigm-agnostic and allows to generate comparatively large datasets with noiseless labels. Source code and data of the novel simulation approach are made available for facilitating its adoption.