Transfer learning in imagined speech EEG-based BCIs

García-Salinas, Jesús S.; Villaseñor-Pineda, Luis; Reyes-García, Carlos A.; Torres-García, Alejandro A.

doi:10.1016/j.bspc.2019.01.006

Cited by 55 publications

(30 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Specifically, TL is applied for improving the performance of the classifier on a new subject (target subject) using the knowledge learnt from a set of different subjects (source subjects). Similar to García-Salinas et al ( 2019 ), the two TL paradigms come under the class of multi-task transfer learning (Evgeniou and Pontil, 2004 ). A deep CNN architecture, similar to the one proposed in Schirrmeister et al ( 2017 ), is used in this work.…”

Section: Feature Extraction and Classificationmentioning

confidence: 94%

“…Transfer learning (TL) is used in García-Salinas et al ( 2019 ) and Cooney et al ( 2019 ) for improving the performance of the classifier. TL is a machine learning approach in which the performance of a classifier in the target domain is improved by incorporating the knowledge learnt from a different domain (Pan and Yang, 2009 ; He and Wu, 2017 ; García-Salinas et al, 2019 ). Specifically in García-Salinas et al ( 2019 ), feature representation transfer is used for representing a new imagined word using the codewords learnt using a set of four other imagined words.…”

Section: Feature Extraction and Classificationmentioning

confidence: 99%

“…TL is a machine learning approach in which the performance of a classifier in the target domain is improved by incorporating the knowledge learnt from a different domain (Pan and Yang, 2009 ; He and Wu, 2017 ; García-Salinas et al, 2019 ). Specifically in García-Salinas et al ( 2019 ), feature representation transfer is used for representing a new imagined word using the codewords learnt using a set of four other imagined words. The codewords were generated using k-means clustering similar to the approach discussed in Plinge et al ( 2014 ) and Lazebnik and Raginsky ( 2008 ).…”

Section: Feature Extraction and Classificationmentioning

confidence: 99%

“…Five studies reviewed in this article makes use of directional prompts. In both García et al ( 2012 ), García-Salinas et al ( 2019 ), five Spanish words, “arriba”, “abajo”, “izquierda”, “derecha”, and “seleccionar” are used as the prompts. These words mean up, down, left,‘right and select, respectively.…”

Section: Feature Extraction and Classificationmentioning

confidence: 99%

See 3 more Smart Citations

Decoding Covert Speech From EEG-A Comprehensive Review

Panachakel

Ramakrishnan

2021

Front. Neurosci.

View full text Add to dashboard Cite

Over the past decade, many researchers have come up with different implementations of systems for decoding covert or imagined speech from EEG (electroencephalogram). They differ from each other in several aspects, from data acquisition to machine learning algorithms, due to which, a comparison between different implementations is often difficult. This review article puts together all the relevant works published in the last decade on decoding imagined speech from EEG into a single framework. Every important aspect of designing such a system, such as selection of words to be imagined, number of electrodes to be recorded, temporal and spatial filtering, feature extraction and classifier are reviewed. This helps a researcher to compare the relative merits and demerits of the different approaches and choose the one that is most optimal. Speech being the most natural form of communication which human beings acquire even without formal education, imagined speech is an ideal choice of prompt for evoking brain activity patterns for a BCI (brain-computer interface) system, although the research on developing real-time (online) speech imagery based BCI systems is still in its infancy. Covert speech based BCI can help people with disabilities to improve their quality of life. It can also be used for covert communication in environments that do not support vocal communication. This paper also discusses some future directions, which will aid the deployment of speech imagery based BCI for practical applications, rather than only for laboratory experiments.

show abstract

Section: Feature Extraction and Classificationmentioning

confidence: 94%

Section: Feature Extraction and Classificationmentioning

confidence: 99%

Section: Feature Extraction and Classificationmentioning

confidence: 99%

Section: Feature Extraction and Classificationmentioning

confidence: 99%

See 2 more Smart Citations

Decoding Covert Speech From EEG-A Comprehensive Review

Panachakel

Ramakrishnan

2021

Front. Neurosci.

View full text Add to dashboard Cite

show abstract

“…The result was a principled way of dealing with outliers in simple data that was, yet achieved the AUC value of 0.75 in cross-subject BCI ( Kadioglu et al, 2018 ). As for the pattern recognition, some algorithms ( Gordon et al, 2017 ; Hajinoroozi et al, 2017 ; García-Salinas et al, 2019 ; Fernandez-Rodriguez et al, 2020 ), such as genetic algorithm and transfer learning ( Huang et al, 2011 , 2017 ; Jalilpour et al, 2020 ), were applied to improve the accuracy and speed of the spellers. On the other hand, the best AUC performance was still not higher than 0.75 ( Krusienski et al, 2006 ; Koçanaoğulları et al, 2020 ), and most works were still based on small samples.…”

Section: Introductionmentioning

confidence: 99%

Improving the Cross-Subject Performance of the ERP-Based Brain–Computer Interface Using Rapid Serial Visual Presentation and Correlation Analysis Rank

Liu

Wang

Sheng

et al. 2020

Front. Hum. Neurosci.

View full text Add to dashboard Cite

The brain-computer interface (BCI) is a system that is designed to provide communication channels to anyone through a computer. Initially, it was suggested to help the disabled, but actually had been proposed a wider range of applications. However, the cross-subject recognition in BCI systems is difficult to break apart from the individual specific characteristics, unsteady characteristics, and environmental specific characteristics, which also makes it difficult to develop highly reliable and highly stable BCI systems. Rapid serial visual presentation (RSVP) is one of the most recent spellers with a clean, unified background and a single stimulus, which may evoke event-related potential (ERP) patterns with less individual difference. In order to build a BCI system that allows new users to use it directly without calibration or with less calibration time, RSVP was employed as evoked paradigm, then correlation analysis rank (CAR) algorithm was proposed to improve the cross-individual classification and simultaneously use as less training data as possible. Fifty-eight subjects took part in the experiments. The flash stimulation time is 200 ms, and the off time is 100 ms. The P300 component was locked to the target representation by time. The results showed that RSVP could evoke more similar ERP patterns among subjects compared with matrix paradigm. Then, the included angle cosine was calculated and counted for averaged ERP waveform between each two subjects. The average matching number of all subjects was 6 for the matrix paradigm, while for the RSVP paradigm, the average matching number range was 20 when the threshold value was set to 0.5, more than three times as much larger, quantificationally indicating that ERP waveforms evoked by the RSVP paradigm produced smaller individual differences, and it is more favorable for crosssubject classification. Information transfer rates (ITR) were also calculated for RSVP and matrix paradigms, and the RSVP paradigm got the average ITR of 43.18 bits/min, which was 13% higher than the matrix paradigm. Then, the receiver operating characteristic (ROC) curve value was computed and compared using the proposed CAR algorithm and traditional random selection. The results showed that the proposed CAR got significantly better performance than the traditional random selection and got the best AUC value

show abstract