Motoaki Kawanabe
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Motoaki Kawanabe has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
- Takeshi Ogawa
Cognitive Mechanisms Laboratories, Advanced Telecommunications Research Institute International, Kyoto, Japan
2 shared publications - Aapo Hyvärinen
Department of Computer Science, University of Helsinki, Helsinki, Finland
1 shared publication - Alexander Binder
ISTD Pillar, Singapore University of Technology and Design, Singapore
1 shared publication - Carmen Vidaurre
Basque Center on Cognition, Brain and Language, Donostia, Spain
1 shared publication - Danny Panknin
Berlin Institute of Technology (TU Berlin), Machine Learning Group, Computer Science, Germany
1 shared publication Frank C. Meinecke
Berlin Institute of Technology (TU Berlin), Machine Learning Group, Computer Science, Germany
1 shared publication
Latest Publications
- Open Access
SPD domain-specific batch normalization to crack interpretable unsupervised domain adaptation in EEG
Electroencephalography (EEG) provides access to neuronal dynamics non-invasively with millisecond resolution, rendering it a viable method in neuroscience and healthcare. However, its utility is limited as current EEG technology does not generalize well across domains (i.e., sessions and subjects) without expensive supervised re-calibration. Contemporary methods cast this transfer learning (TL) problem as a multi-source/-target unsupervised domain adaptation (UDA) problem and address it with deep learning or shallow, Riemannian geometry aware alignment methods. Both directions have, so far, failed to consistently close the performance gap to state-of-the-art domain-specific methods based on tangent space mapping (TSM) on the symmetric positive definite (SPD) manifold. Here, we propose a theory-based machine learning framework that enables, for the first time, learning domain-invariant TSM models in an end-to-end fashion. To achieve this, we propose a new building block for geometric deep learning, which we denote SPD domain-specific momentum batch normalization (SPDDSMBN). A SPDDSMBN layer can transform domain-specific SPD inputs into domain-invariant SPD outputs, and can be readily applied to multi-source/-target and online UDA scenarios. In extensive experiments with 6 diverse EEG brain-computer interface (BCI) datasets, we obtain state-of-the-art performance in inter-session and -subject TL with a simple, intrinsically interpretable network architecture, which we denote TSMNet.
Open AccessSPD domain-specific batch normalization to crack interpretable unsupervised domain adaptation in EEG
Electroencephalography (EEG) provides access to neuronal dynamics non-invasively with millisecond resolution, rendering it a viable method in neuroscience and healthcare. However, its utility is limited as current EEG technology does not generalize well across domains (i.e., sessions and subjects) without expensive supervised re-calibration. Contemporary methods cast this transfer learning (TL) problem as a multi-source/-target unsupervised domain adaptation (UDA) problem and address it with deep learning or shallow, Riemannian geometry aware alignment methods. Both directions have, so far, failed to consistently close the performance gap to state-of-the-art domain-specific methods based on tangent space mapping (TSM) on the symmetric positive definite (SPD) manifold. Here, we propose a theory-based machine learning framework that enables, for the first time, learning domain-invariant TSM models in an end-to-end fashion. To achieve this, we propose a new building block for geometric deep learning, which we denote SPD domain-specific momentum batch normalization (SPDDSMBN). A SPDDSMBN layer can transform domain-specific SPD inputs into domain-invariant SPD outputs, and can be readily applied to multi-source/-target and online UDA scenarios. In extensive experiments with 6 diverse EEG brain-computer interface (BCI) datasets, we obtain state-of-the-art performance in inter-session and -subject TL with a simple, intrinsically interpretable network architecture, which we denote TSMNet.
- Open Access
- 1 February 2022
- Published by Elsevier in NeuroImage
- Vol. 247, 118794
Asymmetric directed functional connectivity within the frontoparietal motor network during motor imagery and execution
- 1 February 2022
- Published by Elsevier in NeuroImage
- Vol. 247, 118794
Open AccessAsymmetric directed functional connectivity within the frontoparietal motor network during motor imagery and execution
- Open Access
Asymmetric directed functional connectivity within the frontoparietal motor network during motor imagery and execution
Both imagery and execution of motor controls consist of interactions within a neuronal network, including frontal motor-related regions and posterior parietal regions. To reveal neural representation in the frontoparietal motor network, two approaches have been proposed thus far: one is decoding of actions/modes related to motor control from the spatial pattern of brain activity; another is to estimate directed functional connectivity, which means a directed association between two brain regions within motor regions. However, directed connectivity among multiple regions of the motor network during motor imagery (MI) or motor execution (ME) has not been investigated. Here, we attempted to characterize the directed functional connectivity within the frontoparietal motor-related networks between the MI and ME conditions. We developed a delayed sequential movement and imagery task to evoke brain activity associated with data under ME and MI via functional magnetic resonance imaging scanning. We applied a causal discovery approach, linear non-Gaussian acyclic causal model, to identify directed functional connectivity among the frontoparietal motor-related brain regions for each condition. We demonstrated higher directed functional connectivity from the contralateral dorsal premotor cortex (dPMC) to the primary motor cortex (M1) in ME than in MI. We mainly identified significant direct effects of the dPMC and ventral premotor cortex (vPMC) to the parietal regions. In particular, connectivity from the dPMC to the superior parietal lobule in the same hemisphere showed significant positive effects across all conditions. Contrastingly, interlateral connectivities from the vPMC to the superior parietal lobule showed significantly negative effects across all conditions. Finally, we found positive effects from A1 to M1 in the same hemisphere, such as the audio-motor pathway. These results indicated that the sources of motor command originated in d/vPMC influenced M1 and parietal regions as achieving ME and MI. Additionally, sequential sounds may functionally facilitate temporal motor processes.
Open AccessAsymmetric directed functional connectivity within the frontoparietal motor network during motor imagery and execution
Both imagery and execution of motor controls consist of interactions within a neuronal network, including frontal motor-related regions and posterior parietal regions. To reveal neural representation in the frontoparietal motor network, two approaches have been proposed thus far: one is decoding of actions/modes related to motor control from the spatial pattern of brain activity; another is to estimate directed functional connectivity, which means a directed association between two brain regions within motor regions. However, directed connectivity among multiple regions of the motor network during motor imagery (MI) or motor execution (ME) has not been investigated. Here, we attempted to characterize the directed functional connectivity within the frontoparietal motor-related networks between the MI and ME conditions. We developed a delayed sequential movement and imagery task to evoke brain activity associated with data under ME and MI via functional magnetic resonance imaging scanning. We applied a causal discovery approach, linear non-Gaussian acyclic causal model, to identify directed functional connectivity among the frontoparietal motor-related brain regions for each condition. We demonstrated higher directed functional connectivity from the contralateral dorsal premotor cortex (dPMC) to the primary motor cortex (M1) in ME than in MI. We mainly identified significant direct effects of the dPMC and ventral premotor cortex (vPMC) to the parietal regions. In particular, connectivity from the dPMC to the superior parietal lobule in the same hemisphere showed significant positive effects across all conditions. Contrastingly, interlateral connectivities from the vPMC to the superior parietal lobule showed significantly negative effects across all conditions. Finally, we found positive effects from A1 to M1 in the same hemisphere, such as the audio-motor pathway. These results indicated that the sources of motor command originated in d/vPMC influenced M1 and parietal regions as achieving ME and MI. Additionally, sequential sounds may functionally facilitate temporal motor processes.
- Open Access
- 1 March 2016
- Published by IOP Publishing in Journal of Physics: Conference Series
- Vol. 699 (1) , 012021
Higher order stationary subspace analysis
Non-stationarity in data is an ubiquitous problem in signal processing. The recent stationary subspace analysis procedure (SSA) has enabled to decompose such data into a stationary subspace and a non-stationary part respectively. Algorithmically only weak non- stationarities could be tackled by SSA. The present paper takes the conceptual step generalizing from the use of first and second moments as in SSA to higher order moments, thus defining the proposed higher order stationary subspace analysis procedure (HOSSA). The paper derives the novel procedure and shows simulations. An obvious trade-off between the necessity of estimating higher moments and the accuracy and robustness with which they can be estimated is observed. In an ideal setting of plenty of data where higher moment information is dominating our novel approach can win against standard SSA. However, with limited data, even though higher moments actually dominate the underlying data, still SSA may arrive on par.
- 1 March 2016
- Published by IOP Publishing in Journal of Physics: Conference Series
- Vol. 699 (1) , 012021
Open AccessHigher order stationary subspace analysis
Non-stationarity in data is an ubiquitous problem in signal processing. The recent stationary subspace analysis procedure (SSA) has enabled to decompose such data into a stationary subspace and a non-stationary part respectively. Algorithmically only weak non- stationarities could be tackled by SSA. The present paper takes the conceptual step generalizing from the use of first and second moments as in SSA to higher order moments, thus defining the proposed higher order stationary subspace analysis procedure (HOSSA). The paper derives the novel procedure and shows simulations. An obvious trade-off between the necessity of estimating higher moments and the accuracy and robustness with which they can be estimated is observed. In an ideal setting of plenty of data where higher moment information is dominating our novel approach can win against standard SSA. However, with limited data, even though higher moments actually dominate the underlying data, still SSA may arrive on par.
- 1 March 2016
- Published by MIT Press in Neural Computation
- Vol. 28 (3) , 445-484
Orthogonal Connectivity Factorization: Interpretable Decomposition of Variability in Correlation Matrices
In many multivariate time series, the correlation structure is nonstationary, that is, it changes over time. The correlation structure may also change as a function of other cofactors, for example, the identity of the subject in biomedical data. A fundamental approach for the analysis of such data is to estimate the correlation structure (connectivities) separately in short time windows or for different subjects and use existing machine learning methods, such as principal component analysis (PCA), to summarize or visualize the changes in connectivity. However, the visualization of such a straightforward PCA is problematic because the ensuing connectivity patterns are much more complex objects than, say, spatial patterns. Here, we develop a new framework for analyzing variability in connectivities using the PCA approach as the starting point. First, we show how to analyze and visualize the principal components of connectivity matrices by a tailor-made rank-two matrix approximation in which we use the outer...
- 1 March 2016
- Published by MIT Press in Neural Computation
- Vol. 28 (3) , 445-484
Orthogonal Connectivity Factorization: Interpretable Decomposition of Variability in Correlation Matrices
In many multivariate time series, the correlation structure is nonstationary, that is, it changes over time. The correlation structure may also change as a function of other cofactors, for example, the identity of the subject in biomedical data. A fundamental approach for the analysis of such data is to estimate the correlation structure (connectivities) separately in short time windows or for different subjects and use existing machine learning methods, such as principal component analysis (PCA), to summarize or visualize the changes in connectivity. However, the visualization of such a straightforward PCA is problematic because the ensuing connectivity patterns are much more complex objects than, say, spatial patterns. Here, we develop a new framework for analyzing variability in connectivities using the PCA approach as the starting point. First, we show how to analyze and visualize the principal components of connectivity matrices by a tailor-made rank-two matrix approximation in which we use the outer...
- Open Access
- 1 January 2015
- Published by Frontiers Media SA in Frontiers in Computational Neuroscience
- Vol. 9
On possibilities and difficulties in adaptive brain-computer interface in real environments
Thanks to the huge efforts by researchers in this field, basic technologies of brain-computer interfaces (BCIs) has been established and practical applications, for instance, BCI rehabilitations are being developed. Based on our previous achievements of non-invasive brain-machine interfaces (BMIs) in experimental rooms, ATR-BICR had conducted the network BMI project (Jul. 2011 – Mar. 2015) in order to develop real-world BMI for supporting elderly and disable people in daily life. For conducting real-world BMI experiments and acquiring daily-life brain activities, we have prepared in the premises of ATR the BMI smart house with various ambient sensors. Our ambitious research proposal gained much attention with expectation and suspicion in the BCI community when the project started. Although we could demonstrate few prototypes of such BMI assistance systems successfully [1], [2], we realized that there exist many remaining issues to achieve our original goal [3]. In this presentation, we will discuss what we can learn from our experiences in the project. One of the important issues is non-stationarity and subject-to-subject variability in brain signals during BCI sessions. Electroencephalographic (EEG) signals are known to be non-stationary and easily affected by artifacts. Since such fluctuations may be caused by changes in the subject's brain processes, e.g. change of task involvement, fatigue, learning effects etc., it is particularly important to alleviate non-stationarity in EEG time series in order to construct useful BCI systems in real-world environments. In this talk, we will briefly explain a few works on this issue developed by international collaborations. One of them is the maxmin Common Spatial Pattern (CSP) in collaboration with the Berlin BCI team, a robust version of the popular spatial filter CSP for EEG-BCI by a maxmin approach [4]. In contrast to standard CSP that maximizes the variance ratio between two conditions based on a single estimate of the class covariance matrices, we propose to robustly compute spatial filters by maximizing the minimum variance ratio within a prefixed set of covariance matrices called the tolerance set. We show that this kind of maxmin optimization makes CSP robust to outliers and reduces its tendency to overfit. Another research direction is the orthogonal connectivity factorization [5] developed in collaboration with Dr. Hyvärinen. In neuroimaging, the correlation structure of multivariate brain signals is non-stationary, i.e. it changes over time and it is now popular to investigate the dynamics of connectivity. Here, we use the PCA approach to short-term connectivity matrices by [6] as the starting point and present new methods to simplify interpretation of the results by finding components in the original data space instead of the connectivity space. Preliminary analysis on magnetoencephalographic (MEG) and EEG data reveal possibly meaningful non-stationarity patterns.
- 1 January 2015
- Published by Frontiers Media SA in Frontiers in Computational Neuroscience
- Vol. 9
Open AccessOn possibilities and difficulties in adaptive brain-computer interface in real environments
Thanks to the huge efforts by researchers in this field, basic technologies of brain-computer interfaces (BCIs) has been established and practical applications, for instance, BCI rehabilitations are being developed. Based on our previous achievements of non-invasive brain-machine interfaces (BMIs) in experimental rooms, ATR-BICR had conducted the network BMI project (Jul. 2011 – Mar. 2015) in order to develop real-world BMI for supporting elderly and disable people in daily life. For conducting real-world BMI experiments and acquiring daily-life brain activities, we have prepared in the premises of ATR the BMI smart house with various ambient sensors. Our ambitious research proposal gained much attention with expectation and suspicion in the BCI community when the project started. Although we could demonstrate few prototypes of such BMI assistance systems successfully [1], [2], we realized that there exist many remaining issues to achieve our original goal [3]. In this presentation, we will discuss what we can learn from our experiences in the project. One of the important issues is non-stationarity and subject-to-subject variability in brain signals during BCI sessions. Electroencephalographic (EEG) signals are known to be non-stationary and easily affected by artifacts. Since such fluctuations may be caused by changes in the subject's brain processes, e.g. change of task involvement, fatigue, learning effects etc., it is particularly important to alleviate non-stationarity in EEG time series in order to construct useful BCI systems in real-world environments. In this talk, we will briefly explain a few works on this issue developed by international collaborations. One of them is the maxmin Common Spatial Pattern (CSP) in collaboration with the Berlin BCI team, a robust version of the popular spatial filter CSP for EEG-BCI by a maxmin approach [4]. In contrast to standard CSP that maximizes the variance ratio between two conditions based on a single estimate of the class covariance matrices, we propose to robustly compute spatial filters by maximizing the minimum variance ratio within a prefixed set of covariance matrices called the tolerance set. We show that this kind of maxmin optimization makes CSP robust to outliers and reduces its tendency to overfit. Another research direction is the orthogonal connectivity factorization [5] developed in collaboration with Dr. Hyvärinen. In neuroimaging, the correlation structure of multivariate brain signals is non-stationary, i.e. it changes over time and it is now popular to investigate the dynamics of connectivity. Here, we use the PCA approach to short-term connectivity matrices by [6] as the starting point and present new methods to simplify interpretation of the results by finding components in the original data space instead of the connectivity space. Preliminary analysis on magnetoencephalographic (MEG) and EEG data reveal possibly meaningful non-stationarity patterns.
- 2 July 2014
- Published by Springer Nature
- p. 211-223
Machine Learning for Visual Concept Recognition and Ranking for Images
- 2 July 2014
- Published by Springer Nature
- p. 211-223
Machine Learning for Visual Concept Recognition and Ranking for Images
- 1 February 2014
- Published by MIT Press in Neural Computation
- Vol. 26 (2) , 349-376
Robust Common Spatial Filters with a Maxmin Approach
Electroencephalographic signals are known to be nonstationary and easily affected by artifacts; therefore, their analysis requires methods that can deal with noise. In this work, we present a way to robustify the popular common spatial patterns (CSP) algorithm under a maxmin approach. In contrast to standard CSP that maximizes the variance ratio between two conditions based on a single estimate of the class covariance matrices, we propose to robustly compute spatial filters by maximizing the minimum variance ratio within a prefixed set of covariance matrices called the tolerance set. We show that this kind of maxmin optimization makes CSP robust to outliers and reduces its tendency to overfit. We also present a data-driven approach to construct a tolerance set that captures the variability of the covariance matrices over time and shows its ability to reduce the nonstationarity of the extracted features and significantly improve classification accuracy. We test the spatial filters derived with this approach and compare them to standard CSP and a state-of-the-art method on a real-world brain-computer interface (BCI) data set in which we expect substantial fluctuations caused by environmental differences. Finally we investigate the advantages and limitations of the maxmin approach with simulations.
- 1 February 2014
- Published by MIT Press in Neural Computation
- Vol. 26 (2) , 349-376
Robust Common Spatial Filters with a Maxmin Approach
Electroencephalographic signals are known to be nonstationary and easily affected by artifacts; therefore, their analysis requires methods that can deal with noise. In this work, we present a way to robustify the popular common spatial patterns (CSP) algorithm under a maxmin approach. In contrast to standard CSP that maximizes the variance ratio between two conditions based on a single estimate of the class covariance matrices, we propose to robustly compute spatial filters by maximizing the minimum variance ratio within a prefixed set of covariance matrices called the tolerance set. We show that this kind of maxmin optimization makes CSP robust to outliers and reduces its tendency to overfit. We also present a data-driven approach to construct a tolerance set that captures the variability of the covariance matrices over time and shows its ability to reduce the nonstationarity of the extracted features and significantly improve classification accuracy. We test the spatial filters derived with this approach and compare them to standard CSP and a state-of-the-art method on a real-world brain-computer interface (BCI) data set in which we expect substantial fluctuations caused by environmental differences. Finally we investigate the advantages and limitations of the maxmin approach with simulations.