Unsupervised Learning for Brain-Computer Interfaces Based on Event-Related Potentials: Review and Online Comparison [Research Frontier]

Hübner, David; Verhoeven, Thibault; Müller, Klaus‐Robert; Kindermans, Pieter-Jan; Tangermann, Michael

doi:10.1109/mci.2018.2807039

Cited by 21 publications

(9 citation statements)

References 61 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Summary of the RPA method Algorithm 1 recapitulates the steps of a classification task using RPA for matching the statistical distributions of the source and target datasets. 2 Re-center the matrices in S and T using (11) and (12), and form new datasets S (rct) and T (rct) = T (rct) ∪ T (rct) u 3 Calculate the ratio of dispersions in S (rct) and T (rct) as in (16) and use it to form the new dataset T (str) = T (str) ∪ T (str) u with matrices as described in (15) 4 Estimate matrices M k and M k for k ∈ {1, . .…”

Section: E Classification On the Transformed Datasetsmentioning

confidence: 99%

“…There has also been works using Bayesian models to describe the variability of the statistics on the source datasets and gather information from multiple datasets [14]. A recent approach that builds upon such Bayesian methods are the works in [15] and [16], which use a special form of the P300 experimental paradigm to do classification with no calibration. There are two main differences between these approaches and our proposal in this work.…”

Section: Introductionmentioning

confidence: 99%

“…Thus, it can be used in addition to an ensembling approach as in [12], which will combine the information from multiple matched-source subjects. Second, our approach is paradigmagnostic and does not rely on any special modification of the experimental setup where the EEG signals are collected (as opposed to [16]), a feature that is appealing to a great number of practionners.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Riemannian Procrustes Analysis: Transfer Learning for Brain–Computer Interfaces

Rodrigues

Jutten

Congedo

2019

IEEE Trans. Biomed. Eng.

156

124

View full text Add to dashboard Cite

This paper presents a Transfer Learning approach for dealing with the statistical variability of EEG signals recorded on different sessions and/or from different subjects. This is a common problem faced by Brain-Computer Interfaces (BCI) and poses a challenge for systems that try to reuse data from previous recordings to avoid a calibration phase for new users or new sessions for the same user. Method: We propose a method based on Procrustes analysis for matching the statistical distributions of two datasets using simple geometrical transformations (translation, scaling and rotation) over the data points. We use symmetric positive definite matrices (SPD) as statistical features for describing the EEG signals, so the geometrical operations on the data points respect the intrinsic geometry of the SPD manifold. Because of its geometry-aware nature, we call our method the Riemannian Procrustes Analysis (RPA). We assess the improvement in Transfer Learning via RPA by performing classification tasks on simulated data and on eight publicly available BCI datasets covering three experimental paradigms (243 subjects in total). Results: Our results show that the classification accuracy with RPA is superior in comparison to other geometry-aware methods proposed in the literature. We also observe improvements in ensemble classification strategies when the statistics of the datasets are matched via RPA. Conclusion and significance: We present a simple yet powerful method for matching the statistical distributions of two datasets, thus paving the way to BCI systems capable of reusing data from previous sessions and avoid the need of a calibration procedure.

show abstract

Section: E Classification On the Transformed Datasetsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Riemannian Procrustes Analysis: Transfer Learning for Brain–Computer Interfaces

Rodrigues

Jutten

Congedo

2019

IEEE Trans. Biomed. Eng.

156

124

View full text Add to dashboard Cite

show abstract

“…BCI speller), it becomes insufficient or even inappropriate when lacking labels and elicitations come from complex and ambiguous inputs. Classes may become unbalanced, and labels absent from the design of the task [47,18]. There should be, however, a minimal amount of found classes involved in particular sensory discrimination under a logical framework, once provided with a standardized presentation of these stimuli [8,7].…”

Section: Hypothesesmentioning

confidence: 99%

A Study Protocol for Randomized Trials on Visual ERP for multiclass discrimination in CAAD applications

Cutellic

Qureshi

2022

Preprint

View full text Add to dashboard Cite

Background: A great part of brain-computer interfaces (BCI’s) research has so far, been focusing on interactions using exogenous responses under selective attention with synchronous and reactive methods for clear relation with controlled stimuli and the widespread applicability of the methods. Over the two past decades, event-related potentials (ERP’s) have become more and more investigated for a broader community of researchers due to the relatively little amount of training necessary for a system to perform and their detection across diverse modalities of acquisition to correlate with the sensory discrimination of dedicated stimuli. The proposed study aims to observe the detection of ERP’s components and correlated neuralphenomena under the visual presentation of complex stimuli and devise processing methods that would generalize their classification for applications in computer-aided architectural design (CAAD), where visual complexity becomes an intrinsic feature of the tasks.Methods: Its objective is exploratory and twofold: evaluating data processing and stimulus presentation methods, as well as the evolution of similar responses in repeated measures intra- and inter-subjects in non-clinical states. The study is divided into 4 phases of offline-online experiments across its timeline. Offline experiments are used to study and validate base methods, and online experiments are used to validate their modifications for application usage. Each of these experiments are framed by cross-sectional and longitudinal sub-studies. The chosen neural phenomena and study, as well as the presentation paradigm, precondition the trial design with repeated measures for averaging temporal waveforms within an exploratory framework. An expected total of at least 200 participants will be recruited over the course of the study. The aim is to recruit for a distribution of age, gender and participants that are involved in studies or practices of architecture, visual arts or related to maximize statistical significance regarding the targeted population segment.Discussion: This study aims to investigate the cardinality of discriminative neural patterns correlated with the presentation of complex visual stimuli by detecting subcomponents of these neural phenomena using the designed system on short and prolonged periods of time and involving participants from architecture, visual arts or related fields. Subsequent results will bring to further discussion the role of visual experience in such system and the range it might address in the population segment.Trial Status: The first phase of the study has been submitted for non-clinical trials and approved by ETH Z¨urich Ethics Commission on the 02nd July 2021 and given the registration ID: EK 2021-N-106. The first trial has been approved on August 1st 2021

show abstract

“…In adaptive classifiers, parameters are incrementally re-estimated and updated over time as new EEG data become available. Unsupervised adaptation methods [Vidaurre et al, 2010, Hübner et al, 2018 are especially useful as they learn from unlabeled data gained from the actual usage of BCIs. Transfer learning techniques [Jayaram et al, 2016, Wu, 2018, Kindermans et al, 2014 use data from the same or similar tasks recorded from current or different subjects to improve the performance.…”

Section: Introductionmentioning

confidence: 99%

Towards physiology-informed data augmentation for EEG-based BCIs

Zlatov¹,

Blankertz²

2022

Preprint

View full text Add to dashboard Cite

Most EEG-based Brain-Computer Interfaces (BCIs) require a considerable amount of training data to calibrate the classification model, owing to the high variability in the EEG data, which manifests itself between participants, but also within participants from session to session (and, of course, from trial to trial). In general, the more complex the model, the more data for training is needed. We suggest a novel technique for augmenting the training data by generating new data from the data set at hand. Different from existing techniques, our method uses backward and forward projection using source localization and a head model to modify the current source dipoles of the model, thereby generating inter-participant variability in a physiologically meaningful way. In this manuscript, we explain the method and show first preliminary results for participant-independent motor-imagery classification. The accuracy was increased when using the proposed method of data augmentation by 13, 6 and 2 percentage points when using a deep neural network, a shallow neural network and LDA, respectively.

show abstract

Unsupervised Learning for Brain-Computer Interfaces Based on Event-Related Potentials: Review and Online Comparison [Research Frontier]

Cited by 21 publications

References 61 publications

Riemannian Procrustes Analysis: Transfer Learning for Brain–Computer Interfaces

Riemannian Procrustes Analysis: Transfer Learning for Brain–Computer Interfaces

A Study Protocol for Randomized Trials on Visual ERP for multiclass discrimination in CAAD applications

Towards physiology-informed data augmentation for EEG-based BCIs

Contact Info

Product

Resources

About