Transfer Learning: A Riemannian Geometry Framework With Applications to Brain–Computer Interfaces

Zanini, Paolo; Congedo, Marco; Jutten, Christian; Said, Salem; Berthoumieu, Yannick

doi:10.1109/tbme.2017.2742541

Cited by 301 publications

(252 citation statements)

References 32 publications

Supporting

Mentioning

249

Contrasting

Order By: Relevance

“…The reader should note that the above estimation approach is applied for each target-conditional distribution individually; hence, if the number of targets is 8 as in the eye movement decoding problem, we need to estimate 8 corresponding linear transformation matrices H, one for each target-conditional distribution. In light of this, from (10) and (11) we observe that the computation of both θ and H involves inverse of the target-conditional covariance matrices. In a limited data scenario, we need to make sure that the covariance matrices are well-conditioned such that they can be adequately inverted.…”

Section: ) Transfer Functionsmentioning

confidence: 99%

Cross-subject decoding of eye movement goals from local field potentials

et al. 2020

View full text Add to dashboard Cite

Objective. We consider the cross-subject decoding problem from local field potential (LFP) signals, where training data collected from the prefrontal cortex (PFC) of a source subject is used to decode intended motor actions in a destination subject. Approach. We propose a novel supervised transfer learning technique, referred to as data centering, which is used to adapt the feature space of the source to the feature space of the destination. The key ingredients of data centering are the transfer functions used to model the deterministic component of the relationship between the source and destination feature spaces. We propose an efficient data-driven estimation approach for linear transfer functions that uses the first and second order moments of the class-conditional distributions. Main result. We apply our data centering technique with linear transfer functions for cross-subject decoding of eye movement intentions in an experiment where two macaque monkeys perform memory-guided visual saccades to one of eight target locations. The results show peak cross-subject decoding performance of 80%, which marks a substantial improvement over random choice decoder. In addition to this, data centering also outperforms standard sampling-based methods in setups with imbalanced training data. Significance. The analyses presented herein demonstrate that the proposed data centering is a viable novel technique for reliable LFP-based cross-subject brain-computer interfacing and neural prostheses.Review of existing work. Despite the challenges imposed by the non-stationary nature of neuronal activity signals, the cross-subject decoding problem in non-invasive, EEG-based BCIs has been addressed and the progress has been reported. To this end, methods from the emerging field of transfer learning have been extensively applied, albeit to varying degrees of success [9]. In its most general, transfer learning refers to a set of data mining and machine learning techniques, procedures and algorithms designed to extrapolate knowledge acquired in a given domain and apply it in different but somewhat related domain [9], [15], [16]. The a priori assumption in transfer learning is that there exist inherent connections, correlations and/or similarities between the domains, and the objective of any transfer learning algorithm is to discover these similarity structures and foster reliable transfer of knowledge across the domains [15]. It should be also noted that transfer learning is often done in a non-parametric, data-driven manner without relying extensively on detailed statistical models; this has proven to be effective in problems characterized with highly non-stationary temporal and/or spatial behaviour.In the context of cross-subject BCIs, which can be seen as one specific example of transfer learning, several techniques for identifying and estimating structural similarities between neurological data collected in different subjects have emerged; they can be organized into several broader categories which we briefly outline and discuss ne...

show abstract

Section: ) Transfer Functionsmentioning

confidence: 99%

Cross-subject decoding of eye movement goals from local field potentials

et al. 2020

View full text Add to dashboard Cite

show abstract

“…See the SM for the complete Table 1(b). We consider only 5 subjects out of the available 9 as in Zanini et al [2018] and Yair et al [2019]. This is because the single-session single-subject classification results on data from each of the remaining 4 subjects were poor, see Ang et al [2012], Barachant et al [2012], Zanini et al [2018].…”

Section: Motor Imagery Taskmentioning

confidence: 99%

“…Table 2 contains the classification precision obtained by Algorithm 1 and the summary of the mean performance of the algorithms. The N\A results were not reported by Zanini et al [2018], since these subjects were classified as "bad" subjects. We observe that Algorithm 1 provides the best results overall.…”

Section: Event Related Potential P300 Taskmentioning

confidence: 99%

Parallel Transport on the Cone Manifold of SPD Matrices for Domain Adaptation

Yair

Ben‐Chen

Talmon

2019

IEEE Trans. Signal Process.

View full text Add to dashboard Cite

The problem of domain adaptation has become central in many applications from a broad range of fields. Recently, it was proposed to use Optimal Transport (OT) to solve it. In this paper, we model the difference between the two domains by a diffeomorphism and use the polar factorization theorem to claim that OT is indeed optimal for domain adaptation in a well-defined sense, up to a volume preserving map. We then focus on the manifold of Symmetric and Positive-Definite (SPD) matrices, whose structure provided a useful context in recent applications. We demonstrate the polar factorization theorem on this manifold. Due to the uniqueness of the weighted Riemannian mean, and by exploiting existing regularized OT algorithms, we formulate a simple algorithm that maps the source domain to the target domain. We test our algorithm on two Brain-Computer Interface (BCI) data sets and observe state of the art performance.Preprint. Under review.

show abstract

“…We can also understand EA as a correction of data shift. If we view each EEG covariance matrix as a point on a Riemannian manifold, then individual differences cause shifts of these points, although they may entail more than just a simple displacement [26]. In order to correct this shift, EA moves the covariance matrices of each subject to center them at the identity matrix.…”

Section: Euclidean Alignment (Ea)mentioning

confidence: 99%

Different Set Domain Adaptation for Brain-Computer Interfaces: A Label Alignment Approach

2020

IEEE Trans. Neural Syst. Rehabil. Eng.

View full text Add to dashboard Cite

A brain-computer interface (BCI) system usually needs a long calibration session for each new subject/task to adjust its parameters, which impedes its transition from the laboratory to real-world applications. Domain adaptation, which leverages labeled data from auxiliary subjects/tasks (source domains), has demonstrated its effectiveness in reducing such calibration effort. Currently, most domain adaptation approaches require the source domains to have the same feature space and label space as the target domain, which limits their applications, as the auxiliary data may have different feature spaces and/or different label spaces. This paper considers different set domain adaptation for BCIs, i.e., the source and target domains have different label spaces. We introduce a practical setting of different label sets for BCIs, and propose a novel label alignment (LA) approach to align the source label space with the target label space. It has three desirable properties: 1) LA only needs as few as one labeled sample from each class of the target subject; 2) LA can be used as a preprocessing step before different feature extraction and classification algorithms; and, 3) LA can be integrated with other domain adaptation approaches to achieve even better performance. Experiments on two motor imagery datasets demonstrated the effectiveness of LA.

show abstract

Transfer Learning: A Riemannian Geometry Framework With Applications to Brain–Computer Interfaces

Abstract: Hence, we make, through the affine transformation proposed, data from different sessions and subject comparable, providing a significant improvement in the BCI transfer learning problem.

Cited by 301 publications

References 32 publications

Cross-subject decoding of eye movement goals from local field potentials

Cross-subject decoding of eye movement goals from local field potentials

Parallel Transport on the Cone Manifold of SPD Matrices for Domain Adaptation

Different Set Domain Adaptation for Brain-Computer Interfaces: A Label Alignment Approach

Contact Info

Product

Resources

About