Int. BCI Meeting Brussels / La Hulpe: Workshop W6, Wed., June 7, 9:30am- 12:30pm, Max Theatre, room Imbuia
Collection of materials and links to talks given, tools presented, software demos etc.
Training up a decoding model based on as few possible training data points as possible is a desirable goal, as it can be pivotal for the usability of a BCI application with patients, for the acceptance by healthy users, or to realize fast adaptations during non-stationary recordings or for transferring between sessions. Our workshop addresses the latest proposed techniques to train classification or regression machine learning models with small datasets, embracing approaches from both, traditional machine learning approaches and deep learning approaches. In addition to talks and discussions, we will have a hands-on programming session in Python to benchmark different classification models.
-
Michael Tangermann, Jordy Thielen, Pierre Guetschel, Matthias Dold, Jan Sosulski Data-driven Neurotechnology lab, Donders Institute, Radboud University, Nijmegen, The Netherlands
-
Joana Pereira Stereotactic and Functional Neurosurgery, University Medical Center Freiburg, Germany
-
Theodore Papadopoulo, Igor Carrara INRIA Sophia Antipolis Méditerranée, Sophia Antipolis Cedex, France
-
Alexandre Gramfort Meta Reality Labs, Paris, France
-
Moritz Grosse-Wentrup Research Group Neuroinformatics, Faculty of Computer Science, University of Vienna, Austria.
-
Sylvain Chevallier Laboratoire Interdisciplinaire des Sciences du Numérique (LISN), Université Paris-Saclay, France
-
Reinmar Kobler Dynamic Brain Imaging, Advanced Telecommunications Research Institute (ATR), Kyoto, Japan RIKEN Center for Advanced Intelligence Project (AIP), Tokyo, Japan
Matthias Dold (presenter), Sebastian Castano, Joana Pereira: Post-hoc EEG labelling for data-efficient benchmarking and model pre-training
Jordy Thielen: Efficient decoding of code-modulated evoked responses.
Michael Tangermann (presenter), Jan Sosulski: Novel sample-efficient classification approaches for ERP data: Time-decoupled LDA, Toeplitz-LDA, UMM
Pierre Guetschel: Embeddings for fast subject-to-subject transfer learning with motor imagery data
Reinmar Kobler: Geometric deep learning to bridge Riemannian transfer learning with end-to-end learning
Igor Carrara, Théodore Papadopoulo (presenter): Toward everyday BCI: Augmented Covariance Method in a reduced dataset setting
Sylvain Chevallier: Geometric transfer learning with PyRiemann (integrating examples from PyRiemann toolbox)
- please add your link here to your toolboxes, demos etc.
- Analysis of code-modulated (visual) evoked potentials: https://neurotechlab.socsci.ru.nl/resources/cvep
- Unsupervised mean-difference maximization toolbox: https://github.com/jsosulski/umm_demo
- Pre-trained neural networks for motor imagery decoding (Pierre Guetschel): https://neurotechlab.socsci.ru.nl/resources/pretrained_imagery_models/
- Geometric transfer learning with PyRiemann: https://pyriemann.readthedocs.io/en/latest/auto_examples/index.html#transfer-learning
- Augmented Covariance Method in a reduced dataset setting: https://github.com/carraraig/ACM-in-Reduced-Dataset-Setting
- Geometric deep learning with TSMNet: https://github.com/rkobler/TSMNet/
- Talk about Post-Hoc Relabeling: https://matthiasdold.github.io/post_hoc_relabeling_talk/
- Notebook for testing post-hoc relabeling with rCSP: https://github.com/bsdlab/BCIC2023_posthoc