Vinay Jayaram

Transfer Learning in Brain-Computer Interfaces Abstract\uFFFDThe performance of brain-computer interfaces (BCIs) improves with the amount of avail

The performance of brain-computer interfaces (BCIs) improves with the amount of available training data; the statistical distribution of this data, however, varies across subjects as well as across sessions within individual subjects, limiting the transferability of training data or trained models between them. In this article, we review current transfer learning techniques in BCIs that exploit shared structure between training data of multiple subjects and/or sessions to increase performance. We then present a framework for transfer learning in the context of BCIs that can be applied to any arbitrary feature space, as well as a novel regression estimation method that is specifically designed for the structure of a system based on the electroencephalogram (EEG). We demonstrate the utility of our framework and method on subject-to-subject transfer in a motor-imagery paradigm as well as on session-to-session transfer in one patient diagnosed with amyotrophic lateral sclerosis (ALS), showing that it is able to outperform other comparable methods on an identical dataset.

Multi-task logistic regression in brain-computer interfaces

A brain-computer interface (BCI) is used to enable communication between humans and machines by decoding elicited brain activity patterns. However, these patterns have been found to vary across subjects or even for the same subject across sessions. Such problems render the performance of a BCI highly specific to subjects, requiring expensive and time-consuming individual calibration sessions to adapt BCI systems to new subjects. This work tackles the aforementioned problem in a Bayesian multi-task learning (MTL) framework to transfer common knowledge across subjects and sessions for the adaptation of a BCI to new subjects. In particular, a recent framework, that is able to exploit the structure of multi-channel electroencephalography (EEG), is extended by a Bayesian hierarchical logistic regression decoder for probabilistic binary classification. The derived model is able to explicitly learn spatial and spectral features, therefore making it further applicable for identification, analysis and evaluation of paradigm characteristics without relying on expert knowledge. An offline experiment with the new decoder shows a significant improvement in performance on calibration-free decoding compared to previous MTL approaches for rule adaptation and uninformed models while also outperforming them as soon as subject-specific data becomes available. We further demonstrate the ability of the model to identify relevant topographies along with signal band-power features that agree with neurophysiological properties of a common sensorimotor rhythm paradigm.

A Transfer Learning Approach for Adaptive Classification in P300 Paradigms

Introduction: The P300 is one of the most widely used brain responses in BCIs today, popularized by none other than the P300 speller itself. However, most systems still require significant subject-specific training to achieve accurate, reliable classification of brain signals. We present an approach to classification that allows for classification with zero subject-specific data and also improves as data is collected. It does this through the use of data from other subjects in order to intelligently regularize the subject-specific solution with a prior over the weight vector. This approach has already been validated on spectral data [1] and so by validating on P300 data as well we show that it is a classification technique that is agnostic to how features are computed from the EEG time series so long as there are multiple subjects or sessions involved. We further introduce a novel method for estimating parameters that drastically reduces the time necessary to implement transfer learning.

A Cognitive Brain-Computer Interface for Patients with Amyotrophic Lateral Sclerosis

Brain-computer interfaces (BCIs) are often based on the control of sensorimotor processes, yet sensorimotor processes are impaired in patients suffering from amyotrophic lateral sclerosis (ALS). We devised a new paradigm that targets higher-level cognitive processes to transmit information from the user to the BCI. We instructed five ALS patients and twelve healthy subjects to either activate self-referential memories or to focus on a process without mnemonic content while recording a high-density electroencephalogram (EEG). Both tasks are designed to modulate activity in the default mode network (DMN) without involving sensorimotor pathways. We find that the two tasks can be distinguished after only one experimental session from the average of the combined bandpower modulations in the theta- (4-7Hz) and alpha-range (8-13Hz), with an average accuracy of 62.5% and 60.8% for healthy subjects and ALS patients, respectively. The spatial weights of the decoding algorithm show a preference for the parietal area, consistent with modulation of neural activity in primary nodes of the DMN.

MOABB: trustworthy algorithm benchmarking for BCIs

OBJECTIVE Brain-computer interface (BCI) algorithm development has long been hampered by two major issues: small sample sets and a lack of reproducibility. We offer a solution to both of these problems via a software suite that streamlines both the issues of finding and preprocessing data in a reliable manner, as well as that of using a consistent interface for machine learning methods. APPROACH By building on recent advances in software for signal analysis implemented in the MNE toolkit, and the unified framework for machine learning offered by the scikit-learn project, we offer a system that can improve BCI algorithm development. This system is fully open-source under the BSD licence and available at https://github.com/NeuroTechX/moabb. MAIN RESULTS We analyze a set of state-of-the-art decoding algorithms across 12 open access datasets, including over 250 subjects. Our results show that even for the best methods, there are datasets which do not show significant improvements, and further that many previously validated methods do not generalize well outside the datasets they were tested on. SIGNIFICANCE Our analysis confirms that BCI algorithms validated on single datasets are not representative, highlighting the need for more robust validation in the machine learning for BCIs community.

Case Series: Slowing Alpha Rhythm in Late-Stage ALS Patients

The alpha peak frequency (APF) of the human electroencephalogram (EEG) is a reliable neurophysiological marker for cognitive abilities. In these case series, we document a shift of the APF towards the lower end of the EEG spectrum in two completely locked-in ALS patients. In not completely locked-in ALS patients, the alpha rhythm lies within the common frequency range. We discuss potential implications of this shift for the largely unknown cognitive state of completely locked-in ALS patients.

Frequency peak features for low-channel classification in motor imagery paradigms

The expansion of brain-computer interfaces (BCIs) to outside the research laboratory has historically been hampered by their difficulty of use. Well-functioning BCIs often require many channels, which can be difficult to properly prepare and require expert support. Low-channel setups, however, can lead to poor or unreliable classification of intent. Here we introduce a novel method for extracting more information from a single EEG channel and test it on a ten subject motor imagery dataset. Instead of looking at bandpower or phase synchrony, we test the average frequency within each trial to see if there are task-dependent changes in the spectral locations of neural frequency peaks. We show that using this feature in combination with standard bandpower features is significantly better than bandpower features alone across subjects, both for standard electrodes and electrodes that include a Laplacian filter.

Bayesian Regression for Artifact Correction in Electroencephalography

Many brain-computer interfaces (BCIs) measure brain activity using electroencephalography (EEG). Unfortunately, EEG is highly sensitive to artifacts originating from non-neural sources, requiring procedures to remove the artifactual contamination from the signal. This work presents a probabilistic interpretation for artifact correction that unifies session transfer of linear models and calibration to upcoming sessions. A linear artifact correction model is derived within a Bayesian multi-task learning (MTL) framework, which captures influences of artifact sources on EEG channels across different sessions to correct for artifacts in new sessions or calibrate with session-specific data. The new model was evaluated with a cross-correlation analysis on a real world EEG data set. We show that the new model matches stateof-the-art correlation reduction abilities, but ultimately converges to a simple group mean model for the experimental data set. This observation leaves the proposed MTL approach open for a more detailed investigations of artifact tasks.

Brain-computer interfacing in amyotrophic lateral sclerosis: Implications of a resting-state EEG analysis.

Despite decades of research on EEG-based brain-computer interfaces (BCIs) in patients with amyotrophic lateral sclerosis (ALS), there is still little known about how the disease affects the electromagnetic field of the brain. This may be one reason for the present failure of EEG-based BCI paradigms for completely locked-in ALS patients. In order to help understand this failure, we have recorded resting state data from six ALS patients and thirty-two healthy controls to investigate for group differences. While similar studies have been attempted in the past, none have used high-density EEG or tried to distinguish between physiological and non-physiological sources of the EEG. We find an ALS-specific global increase in gamma power (30-90 Hz) that is not specific to the motor cortex, suggesting that the mechanism behind ALS affects non-motor cortical regions even in the absence of comorbid cognitive deficits.