2025Activity​​ reportProject-TeamPOTIOC

7.1.1​​​‌ OpenVIBE v3.7.0

Participants: Axel‌ Bouneau, Fabien Lotte‌​‌.

External collaborators: Thomas​​ Prampart [Inria Rennes -​​​‌ Hybrid], Anatole Lécuyer‌ [Inria Rennes - Hybrid]‌​‌.

OpenViBE is an​​ open-source and free software​​​‌ platform dedicated to designing,‌ testing and using of‌​‌ brain-computer interfaces. In 2025,​​ the version 3.7.0 of​​​‌ OpenViBE was released. These‌ updates include, among others,‌​‌ the following features that​​ were developped by the​​​‌ Potioc team:

• Various entropy‌ measures
• Riemannian potato and‌​‌ Riemannian potato field
• EDF​​ file readers/writers

8.1.1 Building a​​​‌ taxonomy of variability factors​ in active BCI

Participants:​‌ Pauline Dreyer, Fabien​​ Lotte.

External collaborators:​​​‌ Raphaëlle Roy [Fédération ENAC​ ISAE-SUPAERO ONERA, Université de​‌ Toulouse, France].

Performance​​ in BCIs is intrinsically​​​‌ shaped by variability, both​ between users and within​‌ the same individual over​​ time. Although recent advances​​​‌ in machine learning have​ provided increasingly sophisticated tools​‌ to mitigate these fluctuations,​​ the underlying factors driving​​​‌ variability in BCI performance​ and EEG features remain​‌ insufficiently understood.

To address​​ this gap, we conducted​​​‌ a comprehensive literature query​ in the PubMed database​‌ using keywords related to​​ BCIs, non-stationarities, and variability​​​‌ in active paradigms, yielding​ an initial corpus of​‌ 177 articles. The analysis​​ revealed a striking imbalance​​​‌ in the literature: a​ large majority of studies​‌ (62.9%) focused on methodological​​ approaches designed to manage​​​‌ variability, while only a​ small proportion explicitly investigated​‌ the factors responsible for​​ it. Based on the​​​‌ reviewed literature, variability factors​ could be broadly grouped​‌ into several categories (see​​ Figure 2).

This​​​‌ distribution highlights the need​ for more studies explicitly​‌ assessing the factors explaining​​ and causing variability in​​​‌ BCI performance and EEG​ features. Identifying and organizing​‌ these factors is essential​​ to improve BCI robustness​​​‌ and interpretability. A structured​ taxonomy would provide a​‌ common ground across disciplines,​​ offering a clear overview​​​‌ of known and emerging​ variability factors, while supporting​‌ the identification of confounding​​ factors, facilitating comparisons across​​​‌ studies, and fostering the​ development of personalized and​‌ adaptive BCI systems. Finally,​​ such a taxonomy could​​​‌ also serve as a​ valuable pedagogical resource for​‌ both newcomers and experts​​ in the field. This​​​‌ work was presented in​ the 11th BCI Meeting​‌ 2025. 34

The image​​ illustrates factors affecting intra-user​​​‌ variability in brain activity​ and physiological states. It​‌ categorizes these factors into​​ several groups: neurophysiological activity,​​​‌ psychological states, physiological activity,​ structural anatomy, individual traits,​‌ and experimental setting. Neurophysiological​​ activity includes changes in​​​‌ brain activity and band​ power fluctuations across different​‌ frequencies. Psychological states refer​​ to temporary mental or​​​‌ emotional states like mood​ and motivation. Physiological activity​‌ covers non-neural signals such​​ as muscle activity and​​​‌ heart rhythm. Structural anatomy​ involves anatomical attributes like​‌ gray matter density. Individual​​ traits encompass demographic and​​​‌ cognitive characteristics. Experimental setting​ includes factors like task​‌ type and environment conditions.​​ (Description generated at January​​​‌ 22nd, 2026 by Albert​ AI with the model​‌ Mistral-Small-3.2-24B)

8.1.2 Investigating Intra-User Variability​​​‌ in mental-imagery BCI: a​ multi-session, multi-context experimental protocol​‌

Participants: Pauline Dreyer,​​ Manon Bourdil, Fabien​​​‌ Lotte.

External collaborators:​ Raphaëlle Roy [Fédération ENAC​‌ ISAE-SUPAERO ONERA, Université de​​ Toulouse, France].

BCI​​​‌ performance and EEG signals​ fluctuate as a result​‌ of interacting sources of​​ variability, including experimental context,​​​‌ time of day, and​ user-related states such as​‌ engagement and fatigue. Although​​ machine learning approaches provide​​​‌ effective tools to manage​ with these fluctuations, the​‌ specific factors influencing BCI​​ performance and EEG features​​ remain insufficiently characterized.

To​​​‌ adress this gap, we‌ conducted a six-month experimental‌​‌ BCI campaign involving 22​​ participants. The protocol (COERLE​​​‌ validation number 2020-32) was‌ designed to capture intra-user‌​‌ variability across time, contexts,​​ and moment of the​​​‌ day. Participants completed multiple‌ sessions distributed over several‌​‌ weeks, performing different mental​​ tasks (motor imagery, mental​​​‌ calculation, and letter/word association)‌ under two different interfaces‌​‌ (see Figure 3):​​ (A) Graz, a low-stimulating​​​‌ context and (B) Brain‌ Hero, a highly-stimulating context‌​‌ with background noise. They​​ also have to complete​​​‌ multiple questionnaires, including assessments‌ of psychological states (e.g.,‌​‌ sleepiness, fatigue) and user​​ experience (e.g., interest, engagement).​​​‌ By combining multiple tasks,‌ interaction contexts, temporal factors,‌​‌ and repeated assessments of​​ psychological states, this protocol​​​‌ provides a comprehensive framework‌ to study the dynamics‌​‌ of intra-user variability in​​ active BCIs. Preliminary results​​​‌ from nine participants who‌ completed all six sessions‌​‌ were presented at the​​ Cortico conference 29 and​​​‌ at the 3rd Edition‌ of the PracticalMEEG workshop.‌​‌

The image consists of​​ two parts labeled A​​​‌ (the Graz context) and‌ B (the BrainHero context).‌​‌ Part A shows a​​ black screen with a​​​‌ math problem "713 -‌ 77 = ?" displayed‌​‌ in the center. Below​​ the problem, there are​​​‌ icons for a hand,‌ maths (with plus, minus,‌​‌ multiply and equals signs),​​ and language with letters​​​‌ A, B, C. The‌ second part of A‌​‌ shows the same screen,​​ but with a blue​​​‌ square - the feedback‌ gauge - appearing in‌​‌ the center lane. Part​​ B shows a colorful​​​‌ game interface with three‌ lanes, each containing a‌​‌ circular target. The left​​ image shows a hand​​​‌ icon with lanes for‌ a hand, maths (with‌​‌ plus, minus, multiply, equals),​​ and letters A, B,​​​‌ C. The right image‌ shows a similar setup,‌​‌ but with a white​​ line pointing to the​​​‌ center lane containing a‌ green square, representing the‌​‌ feedback gauge.

8.1.3 Investigating variabilities​​ in motor-imagery BCI in​​​‌ the NEARBY project: protocol‌ and data acquisition

Participants:‌​‌ Juliette Meunier, Simon​​ Kojima, Fabien Lotte​​​‌, Sébastien Rimbert.‌

External collaborators: Maurice Rekrut‌​‌ [DFKI, Saarbrucken, Germany],​​ Marc Tabie [DFKI, Bremen,​​​‌ Germany], Niklas Küper‌ [DFKI, Bremen, Germany],‌​‌ Benedikt Wirth [DFKI, Saarbrucken,​​ Germany].

The NEARBY​​​‌ (Noise and Variability-free BCI‌ Systems for Out-of-the-lab Use)‌​‌ project aimed to study​​ the variability of BCI​​​‌ according to different factors,‌ notably to understand the‌​‌ variability between or within​​ users, which could explain​​​‌ the poor reliability of‌ BCI. Therefore, the project‌​‌ aimed to (1) collect​​ an extensive database of​​​‌ EEG under different conditions‌ and types of BCIs‌​‌ in order (2) identify​​ factors of variability explaining​​​‌ BCI performances and (3)‌ develop new machine learning‌​‌ algorithms which take into​​ account such factors of​​​‌ variability and/or be robust‌ to them. In this‌​‌ context, we are collaborating​​ with DFKI (German Research​​​‌ Center for Artificial Intelligence)‌ in order to collect‌​‌ data of Speech and​​​‌ Motor Imagery (SI/MI) BCIs.​ Our objective in Bordeaux​‌ is to gather data​​ from MI BCI. To​​​‌ achieve this, we have​ established a protocol for​‌ MI BCI that incorporates​​ various factors of variability.​​​‌

This protocol aims to​ collect data from 16​‌ participants over 12 BCI​​ training sessions for each​​​‌ of them. During each​ session, participants are asked​‌ to perform motor imagery​​ tasks involving both feet,​​​‌ their non-dominant hand, as​ well as a resting​‌ condition. Questionnaires are administered​​ before the first session​​​‌ and during each session​ in order to investigate​‌ both inter- and intra-individual​​ factors. To date, data​​​‌ from six participants have​ been collected.

8.1.4 New​‌ Metrics of Event-Related (De)Synchronization​​ Temporal Variability

Participants: Simon​​​‌ Kojima, Fabien Lotte​.

Motor Imagery-based (MI)​‌ Brain-Computer Interface (BCI) detect​​ imagined limb movements from​​​‌ electroEncephaloGraphy (EEG) to translate​ them into commands for​‌ various applications. They do​​ so by analyzing sensorimotor​​​‌ EEG rhythms, typically event-related​ (de)synchronization (ERD/S) over the​‌ motor cortex. Despite MI​​ task intuitiveness and their​​​‌ many BCI applications, not​ all users achieve sufficient​‌ MI classification accuracy, notably​​ due to large intra-​​​‌ and inter-user variability in​ ERD/S. Understanding this variability​‌ is thus crucial for​​ finding ways to enhance​​​‌ BCI classification performance, but​ BCI variability metrics are​‌ lacking. Therefore, our work​​ proposes two new ERD/S​​​‌ variability metrics and studies,​ on a large MI-BCI​‌ dataset ($N=85$ users), how these​​​‌ and two existing metrics​ can explain BCI performance.​‌ Results show that temporal​​ variability of ERD/S—both within​​​‌ and across trials—negatively correlates​ ($r=-‌0.28$ to​​ $-0.\mathrm{34‌}$) with BCI performance​ in the within-user scenario​‌ (with a user-specific classifier).​​ In the cross-users scenario​​​‌ (with a generic cross-user​ classifier), test users variability​‌ metrics, including ERD/S temporal​​ and amplitude variability, were​​​‌ negatively correlated with performance​ ($r=-‌0.30$ to​​ $-0.\mathrm{39‌}$). These findings demonstrate​ the value of metrics​‌ to quantify ERD/S variability.​​ They may also guide​​​‌ future design strategies for​ BCI user training or​‌ machine learning. This work​​ was published as a​​​‌ preprint in HAL 33​.

Bar plots showing​‌ the correlation between each​​ variability metric and classification​​​‌ accuracy. Left: within-user classification​ scenario. Right: cross-user classification​‌ scenario.

8.1.5 Gender Influence on​​ Motor Imagery BCI Performance​​​‌

Participants: Simon Kojima,​ Fabien Lotte.

In​‌ Motor Imagery BCI research,​​ the influence of the​​​‌ user's biological gender on​ BCI performance has not​‌ yet been sufficiently investigated.​​ In our previous work,​​​‌ no significant gender-related differences​ in mu-band ERD amplitude​‌ were observed; however, the​​ impact of gender on​​​‌ BCI performance itself was​ not examined.

In this​‌ study, we investigated how​​ the gender proportion of​​​‌ the training data affects​ BCI performance in a​‌ cross-user MI-BCI classification task.​​ For each test user,​​ the training set size​​​‌ was fixed at 40‌ users, while the proportion‌​‌ of female users was​​ varied across five levels​​​‌ (0.0, 0.25, 0.5, 0.75,‌ and 1.0) through random‌​‌ sampling. This procedure was​​ applied to 86 test​​​‌ users, and changes in‌ classification accuracy were evaluated‌​‌ using a Riemannian classifier​​ (tangent space + logistic​​​‌ regression).

The relationship between‌ the proportion of female‌​‌ users in the training​​ set and classification accuracy​​​‌ was analyzed using a‌ repeated-measures correlation. For female‌​‌ test users, a moderate​​ and statistically significant positive​​​‌ correlation was observed (‌$r=0.‌‌33$, $p=0.000033$).​​​‌ In contrast, no significant‌ correlation was found for‌​‌ male test users (​​$r=0.‌10$, $p=‌0.18$).‌​‌

These results suggest that,​​ while the influence of​​​‌ the gender composition of‌ the training data on‌​‌ classification accuracy is limited​​ for male test users,​​​‌ the proportion of female‌ users in the training‌​‌ data may affect classification​​ performance for female test​​​‌ users. Overall, this study‌ indicates that machine learning‌​‌ models in cross-user MI-BCI​​ systems may exhibit behavior​​​‌ that reflects the gender‌ composition of the training‌​‌ data. Future work should​​ validate these findings using​​​‌ multiple datasets and classification‌ models.

Bar plots showing‌​‌ the classification accuracies when​​ the gender proportion in​​​‌ the training set was‌ changed. Left: Female test‌​‌ users. Right: Male test​​ users.

8.1.6 Quantifying Inter-and​​ Intra-Subject Variability of Sensorimotor​​​‌ Desynchronization Induced by Median‌ Nerve Stimulation and Motor‌​‌ Imagery for BCI

Participants:​​ Valérie Marissens Cueva,​​​‌ Fabien Lotte, Sébastien‌ Rimbert.

External collaborators:‌​‌ Laurent Bougrain [LORIA, Paris​​ Brain Institute].

Motor​​​‌ Imagery-based Brain-Computer Interfaces (MI-BCIs)‌ enable users to control‌​‌ external devices by interpreting​​ sensorimotor activity recorded via​​​‌ ElectroEncephaloGraphy (EEG). Median Nerve‌ Stimulation (MNS) has recently‌​‌ emerged as a promising​​ alternative motor task for​​​‌ BCI applications. However, intra-and‌ inter-subject EEG variability remains‌​‌ a major challenge, affecting​​ BCI system reliability. While​​​‌ variability is a well-known‌ issue, its precise sources‌​‌ and impact on different​​ EEG patterns remain unclear,​​​‌ with a lack of‌ formal and quantitative studies‌​‌ of BCI variability. Thus,​​ this study quantifies intra-and​​​‌ inter-subject variability in MNS-induced‌ sensorimotor desynchronization (ERD) and‌​‌ compares it with that​​ of MI. Results show​​​‌ that MI elicits stronger‌ ERD with lower intra-subject‌​‌ variability, suggesting more consistent​​ activation patterns, while inter-subject​​​‌ variability is similar between‌ tasks. Additionally, the variability‌​‌ of classification accuracies based​​ on Riemannian geometry exhibits​​​‌ a similar trend. These‌ findings provide insights into‌​‌ EEG variability and its​​ implications for BCI design.​​​‌ Identifying stable neural patterns‌ could improve MI-and MNS-based‌​‌ BCIs, particularly for applications​​ such as intraoperative awareness​​​‌ monitoring 21.

8.2.1 Geometry-Aware visualization​‌ of high dimensional Symmetric​​ Positive Definite matrices

Participants:​​​‌ Fabien Lotte.

External​ collaborators: Thibault de Surrel​‌ [LAMSADE], Sylvain Chevallier​​ [Univ. Paris-Saclay], Florian​​​‌ Yger [INSA Rouen].​

Symmetric Positive Definite (SPD)​‌ matrices are pervasive in​​ machine learning, from data​​​‌ features (such as covariance​ matrices) to optimization process,​‌ notably to represent EEG​​ signals in BCI. These​​​‌ matrices induce a Riemannian​ structure, where the curvature​‌ plays a critical role​​ in the success of​​​‌ approaches based on those​ geometries. Yet, for ML​‌ practitioners wanting to visualize​​ SPD matrices, the existing​​​‌ (flat) Euclidean approaches will​ hide the curvature of​‌ the manifold. To overcome​​ this lack of expressivity​​​‌ in the existing algorithms,​ we introduced Riemannian versions​‌ of two state-of-the-art techniques,​​ namely t-SNE and Multidimensional​​​‌ Scaling. Therefore, we are​ able to reduce a​‌ set of $c\times c$ SPD matrices into​​​‌ a set of $\mathrm{2}\times 2$ SPD matrices​‌ in order to capture​​ the curvature information and​​​‌ avoid any distortion induced​ by flattening the representation​‌ in a Euclidean setup.​​ Moreover, our approaches pave​​​‌ the way for targeting​ more general dimensionality reduction​‌ applications while preserving the​​ geometry of the data.​​​‌ We performed experiments on​ controlled synthetic dataset to​‌ ensure that the low-dimensional​​ representation preserves the geometric​​​‌ properties of both SPD​ Gaussian and geodesics. We​‌ also conducted experiments on​​ various real datasets, such​​​‌ as video, anomaly detection,​ EEG signal and others.​‌ Results indicate that our​​ dimensionality reduction methods that​​​‌ are geometry-aware lead to​ better - more accurate​‌ dimensionality reduction than their​​ euclidean counterparts. This work​​​‌ was published in the​ Transactions on Machine Learning​‌ Research (TMLR) journal 18​​.

The image contains​​​‌ six 3D scatter plots​ visualizing data using different​‌ dimensionality reduction techniques. Each​​ plot distinguishes between left-hand​​​‌ and right-hand data, and​ between two sessions. The​‌ techniques include Riemannian t-SNE,​​ Riemannian MDS, UMAP, Euclidean​​​‌ t-SNE, Euclidean MDS, and​ PGA. Each scatter plot​‌ uses color (blue for​​ left-hand, green for right-hand)​​​‌ and shape (circles for​ session 1, diamonds for​‌ session 2) to differentiate​​ the data points. The​​​‌ plots illustrate how these​ techniques cluster and separate​‌ the data in three-dimensional​​ space. (Description generated at​​​‌ January 22nd, 2026 by​ Albert AI with the​‌ model Mistral-Small-3.2-24B)

8.2.2 Wrapped​​ Gaussian on the manifold​​​‌ of Symmetric Positive Definite​ Matrices

Participants: Fabien Lotte​‌.

External collaborators: Thibault​​ de Surrel [LAMSADE],​​​‌ Sylvain Chevallier [Univ. Paris-Saclay]​, Florian Yger [INSA​‌ Rouen].

Circular and​​ non-flat data distributions are​​​‌ prevalent across diverse domains​ of data science, yet​‌ their specific geometric structures​​ often remain underutilized in​​ machine learning frameworks. A​​​‌ principled approach to accounting‌ for the underlying geometry‌​‌ of such data is​​ pivotal, particularly when extending​​​‌ statistical models, like the‌ pervasive Gaussian distribution. In‌​‌ this work, we tackled​​ those issue by focusing​​​‌ on the manifold of‌ symmetric positive definite (SPD)‌​‌ matrices, a key focus​​ in information geometry in​​​‌ general and in current‌ BCI work in particular.‌​‌ We introduced a non-isotropic​​ wrapped Gaussian by leveraging​​​‌ the exponential map, we‌ derive theoretical properties of‌​‌ this distribution and propose​​ a maximum likelihood framework​​​‌ for parameter estimation. Furthermore,‌ we reinterpret established classifiers‌​‌ on SPD through a​​ probabilistic lens and introduce​​​‌ new classifiers based on‌ the wrapped Gaussian model.‌​‌ Experiments on synthetic and​​ real-world datasets (including EEG-BCI​​​‌ data sets) demonstrate the‌ robustness and flexibility of‌​‌ this geometry-aware distribution, underscoring​​ its potential to advance​​​‌ manifold-based data analysis. This‌ work lays the groundwork‌​‌ for extending classical machine​​ learning and statistical methods​​​‌ to more complex and‌ structured data. This work‌​‌ was published in the​​ ICML conference 23.​​​‌ Such Wrapped Gaussian were‌ then used to reintrepet‌​‌ probabilistically various BCI machine​​ learning algorithms on SPD​​​‌ matrices. This was published‌ in the GSI conference‌​‌ 24, where this​​ work was distinguished in​​​‌ the official "selected papers"‌ list of the conference.‌​‌

The image depicts a​​ mathematical concept related to​​​‌ statistics and differential geometry.‌ It shows two spaces:‌​‌ a manifold (labeled as​​ Pd) and a Euclidean​​​‌ space (labeled as TpPd).‌ In the Euclidean space,‌​‌ there's a normal distribution​​ represented as N(μ, Σ).​​​‌ A point 0n in‌ the Euclidean space is‌​‌ connected to a point​​ p in the manifold​​​‌ by a dotted vertical‌ line. The mapping between‌​‌ these spaces is illustrated​​ using functions Logp and​​​‌ Expp, which translate points‌ between the Euclidean space‌​‌ and the manifold. The​​ green region represents the​​​‌ Wrapped Gaussian distribution WG(p;‌ μ, Σ) on the‌​‌ manifold, corresponding to the​​ standard Gaussian distribution on​​​‌ the Euclidean space. (Description‌ generated partially at January‌​‌ 22nd, 2026 by Albert​​ AI with the model​​​‌ Mistral-Small-3.2-24B)

8.2.3 BCI classifiers​​ integrating measures of variability​​​‌ factors: a mini-review

Participants:‌ David Trocellier, Fabien‌​‌ Lotte.

External collaborators:​​ Bernard Nkaoua [Bordeaux population​​​‌ Health /Univ. Bordeaux].‌

BCIs are sensitive to‌​‌ variability factors, such as​​ changes in mental states,​​​‌ experimental setups, and individual‌ neurophysiological differences, which degrade‌​‌ classification performance. We proposed​​ a state-of-the-art review which​​​‌ examines machine learning approaches‌ in BCI that integrate‌​‌ variability factors to enhance​​ robustness and classification performance​​​‌ 25. We conducted‌ a PRISMA review, finally‌​‌ identifying nine relevant papers​​ and we proposed a​​​‌ taxonomy based on variability‌ factors and their integration‌​‌ methods. While promising results,​​ such as improved classification​​​‌ accuracy and feature separability,‌ were observed, more research‌​‌ needs to be done​​ to better understand the​​​‌ interaction between classifiers and‌ variability factors and how‌​‌ it can enhance algorithmic​​ robustness.

8.2.4 Guidelines, tutorials​​​‌ and reflexions on AI/machine‌ learning use in BCI‌​‌

Participants: Fabien Lotte,​​​‌ Marc Welter, David​ Trocellier, Sebastien Rimbert​‌, Pauline Dreyer.​​

External collaborators: David Carlson​​​‌ [Duke University School of​ Medicine, USA], Ricardo​‌ Chavarriaga [ZHAW, Switzerland],​​ Yiling Liu [Duke University​​​‌ School of Medicine, USA]​, Bao-Liang Lu [Shanghai​‌ Jiao Tong University, China]​​, Tommaso Dorigo [INFN:​​​‌ Padova, Italy], Stephanie​ Cernera [UC San Francisco,​‌ USA].

ML ability​​ to capture intricate patterns​​​‌ makes it vital in​ neural engineering research. With​‌ its increasing use, ensuring​​ the validity and reproducibility​​​‌ of ML methods is​ critical. Unfortunately, this has​‌ not always been the​​ case in practice, as​​​‌ there have been recent​ retractions across various scientific​‌ fields due to the​​ misuse of ML methods​​​‌ and validation procedures. To​ address these concerns, we​‌ propose the first version​​ of the neural engineering​​​‌ reproducibility and validity essentials​ for ML (NERVE-ML) checklist,​‌ a framework designed to​​ promote the transparent, reproducible,​​​‌ and valid application of​ ML in neural engineering​‌ 13. We highlight​​ some of the unique​​​‌ challenges of model validation​ in neural engineering, including​‌ the difficulties from limited​​ subject numbers, repeated or​​​‌ non-independent samples, and high​ subject heterogeneity. Through detailed​‌ case studies, we demonstrate​​ how different validation approaches​​​‌ can lead to divergent​ scientific conclusions, highlighting the​‌ importance of selecting appropriate​​ procedures guided by the​​​‌ NERVE-ML checklist. Effectively addressing​ these challenges and properly​‌ scoping scientific conclusions will​​ ensure that ML contributes​​​‌ to, rather than hinders,​ progress in neural engineering.​‌ Our case studies demonstrate​​ that improper validation approaches​​​‌ can result in flawed​ studies or overclaimed scientific​‌ conclusions, complicating the scientific​​ discourse. The NERVE-ML checklist​​​‌ effectively addresses these concerns​ by providing guidelines to​‌ ensure that ML approaches​​ in neural engineering are​​​‌ reproducible and lead to​ valid scientific conclusions. By​‌ effectively addressing these challenges​​ and properly scoping scientific​​​‌ conclusions guided by the​ NERVE-ML checklist, we aim​‌ to help pave the​​ way for a future​​​‌ where ML reliably enhances​ the quality and impact​‌ of neural engineering research.​​ Beyond such checklist, we​​​‌ also proposed various tutorials​ on BCI research 14​‌ and BCI design 31​​, as well as​​​‌ reflexions on the use​ of AI/ML in various​‌ scientific fields, including in​​ BCI research 15.​​​‌

8.3.1 An​​​‌ Active Inference perspective on​ Neurofeedback/BCI Training

Participants: Come​‌ Annicchiarico, Fabien Lotte​​.

External collaborators: Jérémie​​​‌ Mattout [CNRL / Inserm,​ Lyon, France].

Neurofeedback​‌ training (NFT) and BCI​​ training both aim to​​​‌ teach self-regulation of brain​ activity through real-time feedback,​‌ but suffers from highly​​ variable outcomes and poorly​​​‌ understood mechanisms, hampering its​ validation. To address these​‌ issues, we propose a​​ formal computational model of​​ the NFT/BCI closed loop.​​​‌ Using Active Inference, a‌ Bayesian framework modelling perception,‌​‌ action, and learning, we​​ simulate agents interacting with​​​‌ an NFT/BCI environment. This‌ enables us to test‌​‌ the impact of design​​ choices (e.g., feedback quality,​​​‌ biomarker validity) and subject‌ factors (e.g., prior beliefs‌​‌ - linked to instructions​​ received) on training. Simulations​​​‌ show that training effectiveness‌ is sensitive to feedback‌​‌ noise or bias, and​​ to prior beliefs (highlighting​​​‌ the importance of guiding‌ instructions), but also reveal‌​‌ that perfect feedback is​​ insufficient to guarantee high​​​‌ performance. This approach provides‌ a tool for assessing‌​‌ and predicting NFT variability,​​ interpret empirical data, and​​​‌ potentially develop personalized training‌ protocols 32.

8.3.2‌​‌ Sense-IT project

Participants: Sébastien​​ Rimbert, Stéphanie Fleck​​​‌.

External collaborators: Mathilde‌ Yousefi [Laboratoire PErSEUs],‌​‌ Jérémy Frey [Qualya],​​ Altamira Gabriela Herrera [Capgemini]​​​‌.

Kinesthetic motor imagery‌ (KMI) combined with BCI-based‌​‌ neurofeedback is a promising​​ approach for post-stroke upper-limb​​​‌ rehabilitation, but its effectiveness‌ is often limited by‌​‌ insufficient feedback. The Sense-IT​​ project explores sensorimotor neurofeedback​​​‌ by combining a deformable‌ tangible interface with a‌​‌ gamified BCI providing visual​​ and kinesthetic feedback.

A​​​‌ double-blind mixed study (N=36)‌ compared visual, kinesthetic, and‌​‌ bimodal feedback modalities. While​​ no significant differences were​​​‌ observed in motor cortex‌ activation, subjective results showed‌​‌ higher engagement, better task​​ understanding, and improved perceived​​​‌ control with kinesthetic and‌ multimodal feedback.

These findings‌​‌ emphasize the value of​​ embodied and multisensory feedback​​​‌ to improve user experience‌ and acceptability in BCI-based‌​‌ motor rehabilitation 28.​​

Sense-IT is a deformable​​​‌ TUI used as a‌ BCI output device to‌​‌ help the user perceive​​ their motor brain activity.​​​‌ Deforming enables the opening‌ and/or closing of the‌​‌ hand, raising and closing​​ of the thumb, to​​​‌ reproduce grasp, pinch, and‌ hand-opening movements. It aims‌​‌ to aid the neurostimulation​​ of motor cerebral areas​​​‌ damaged by stroke, resulting‌ in hemiplegia or hemiparesis‌​‌ of the upper limbs.​​

8.3.3 STIM-BCI‌​‌ project

Participants: Loïc Bechon​​, Sebastien Rimbert,​​​‌ Fabien Lotte.

External‌ collaborators: Stéphanie Fleck [Laboratoire‌​‌ Perseus / Univ. Loraine]​​.

One of the​​​‌ most prominent BCI interaction‌ paradigms is motor imagery‌​‌ (MI)-based BCI. However, issues​​ such as BCI inefficiency,​​​‌ intra- and inter-subject variability,‌ and laboratory-bound constraints must‌​‌ be addressed for broader​​ usability. A promising approach​​​‌ to overcome these limitations‌ is the use of‌​‌ external somatosensory stimulation, such​​ as vibrotactile or median​​​‌ nerve stimulation (MNS), to‌ enhance both performance and‌​‌ usability. MNS already showed​​ 12% higher accuracy than​​​‌ MI-BCIs without MNS. The‌ project aims to investigate‌​‌ the stimulation patterns, classification​​ methods, and user acceptability​​​‌ of such devices in‌ order to develop an‌​‌ effective somatosensory stimulation–augmented BCI.​​ Three experimental protocols have​​​‌ been designed to examine‌ the effects of different‌​‌ MNS parameters (i.e., intensity,​​ duration, and laterality). Part​​​‌ of this work was‌ presented at CORTICO 2025‌​‌ 19. These protocols​​ are currently being ran,​​​‌ in the data collection‌ phase. Additional ongoing work‌​‌ includes a systematic review​​​‌ of somatosensory stimulation–based BCI​ to explore others potentials​‌ stimulations to investigate during​​ the project.

8.3.4 Evolution​​​‌ of users' subjective experience​ over three training sessions​‌ with an EEG Motor-Imagery​​ Brain-Computer Interface (MI-BCI)

Motor​​​‌ Imagery-based Brain-Computer Interfaces (MI-BCIs)​ have been shown to​‌ be promising for numerous​​ applications, including sport training​​​‌ and entertainment for healthy​ users, but also for​‌ improving or restoring functions​​ in neurological and neuropsychiatric​​​‌ disorders, e.g., for motor​ rehabilitation post-stroke or for​‌ attention training in attention​​ deficits. Reliable interactions with​​​‌ such MI-BCIs require a​ heavy training process for​‌ both the machine and​​ the user. Yet, how​​​‌ User eXperience (UX) evolves​ during standard training is​‌ still largely unclear, both​​ within and between sessions/days.​​​‌ Through an exploratory study,​ we investigated the variations​‌ of users' answers to​​ a UX questionnaire when​​​‌ training with a standard​ left vs. right-hand MI-BCI​‌ 17. 24 healthy​​ novice users engaged in​​​‌ 3 training sessions (with​ 12 runs each) on​‌ different days. Each short​​ run was followed by​​​‌ six questions on screen​ measuring UX factors on​‌ scales from 1 to​​ 10: mental demand, performance,​​​‌ mental effort, frustration, mental​ fatigue and anxiety. Interestingly,​‌ BCI performances did not​​ correlate with any subjective​​​‌ UX measure in this​ study. However, a time​‌ effect was observed. Within​​ session, the results suggested​​​‌ that mental demand, effort,​ and fatigue significantly augmented​‌ during BCI operation, and​​ that frustration significantly fluctuated​​​‌ but did not differ​ pre-vs. post-session (see Figure​‌ 9). Between sessions,​​ the first session was​​​‌ rated significantly more challenging​ than the other two​‌ regarding frustration, anxiety, mental​​ demand, mental effort and​​​‌ mental fatigue. This highlights​ the importance of conducting​‌ studies across sessions and​​ of considering the users'​​​‌ mental states during BCI​ use, for improving UX​‌ and thus possibly BCI​​ treatment outcome.

The image​​​‌ displays six rows of​ statistical graphs, each containing​‌ two plots: a scatter​​ plot on the left​​​‌ and a box plot​ on the right. The​‌ scatter plots show data​​ points for different conditions​​​‌ across three x-axis categories​ labeled SL, S2, and​‌ S3. The box plots​​ compare data distributions across​​​‌ four groups labeled 1,​ 2, 3, and 4.​‌ Each row represents a​​ different measured variable: task​​​‌ performance, mental fatigue, mental​ effort, mental demand, frustration,​‌ and anxiety. The scatter​​ plots depict individual data​​​‌ points with error bars,​ while the box plots​‌ show median values, quartiles,​​ and outliers. Overall, the​​​‌ image provides a comparative​ analysis of these variables​‌ under different conditions and​​ groups. (Description generated at​​​‌ January 23rd, 2026 by​ Albert AI with the​‌ model Mistral-Small-3.2-24B)

8.4.1 Reliable predictor of​ BCI motor imagery performance​‌ using median nerve stimulation​​

Participants: Valérie Marissens Cueva​​​‌, Fabien Lotte,​ Sébastien Rimbert.

External​‌ collaborators: Laurent Bougrain [LORIA,​​ Paris Brain Institute],​​​‌ Camilla Mannino [Inria Paris​ / ICM], Marie-Constance​‌ Corsi [Inria Paris /​​ ICM].

Approximately 30%​​ of individuals fail to​​​‌ effectively use a Brain-Computer‌ Interface (BCI), a phenomenon‌​‌ known as BCI deficiency.​​ Predicting BCI performance is​​​‌ thus crucial for optimizing‌ system parameters, selecting users,‌​‌ and harmonizing participant groups.​​ While BCI performance prediction​​​‌ based on motor imagery‌ (MI) remains an open‌​‌ question, various neurophysiological predictors​​ assess motor cortex activation​​​‌ ability. We propose a‌ novel predictor based on‌​‌ Median Nerve Stimulation (MNS),​​ specifically, the minimum value​​​‌ (200–800 ms post-MNS) of‌ the Event-Related Desynchronization (ERD)‌​‌ at electrode C3 using​​ a small Laplacian filter.​​​‌ Right-hand MI vs. rest‌ BCI performance was evaluated‌​‌ offline using a Tangent​​ Space Logistic Regression classifier​​​‌ in 31 subjects. BCI‌ accuracy strongly correlated with‌​‌ post-MNS ERD (Spearman’s rho​​ = -0.71, p <​​​‌ 0.001) 16. Beyond‌ correlation analysis, we actually‌​‌ predicted BCI performance using​​ a Least Absolute Shrinkage​​​‌ and Selection Operator (LASSO)‌ regression model, trained on‌​‌ six MNS-based features: minimum​​ ERD (200–800 ms) and​​​‌ maximum ERS (800–1500 ms)‌ post-MNS in mu, beta,‌​‌ and mu+beta. Using only​​ these features, LASSO predicted​​​‌ MI-BCI accuracies with a‌ correlation of rho =‌​‌ 0.65 (p < 0.01)​​ between real and predicted​​​‌ accuracies. We also tested‌ whether the three post-MNS‌​‌ ERD could predict a​​ performance group, rather than​​​‌ the exact accuracy score‌ 20. LASSO achieved‌​‌ 74.19% accuracy for two​​ groups, though performance decreased​​​‌ to 45.16% for three‌ groups. Based on reports‌​‌ from the literature, our​​ new MNS based predictor​​​‌ seems to outperform state-of-the-art‌ alternatives, including SMR and‌​‌ MeanSP (rho = 0.53),​​ PPfactor (rho = 0.48),​​​‌ and Spectral Entropy (rho‌ = 0.65). These results‌​‌ suggest an inherent neurophysiological​​ predisposition for MI-BCI success.​​​‌ Future work will integrate‌ multiple predictors into a‌​‌ single model for improved​​ accuracy 30.

The​​​‌ image shows two sets‌ of topographical head maps‌​‌ (3-groups and 2-groups) comparing​​ MI and MNS conditions​​​‌ under low, medium, and‌ high classification performance levels.‌​‌ Each map uses a​​ color scale from blue​​​‌ (-2 dB) to red‌ (2 dB) to represent‌​‌ different signal strengths. The​​ left side shows three​​​‌ rows (low, medium, high)‌ for each condition (MI‌​‌ and MNS) under 3-groups.​​ The right side shows​​​‌ two columns (low and‌ high) for each condition‌​‌ (MI and MNS) under​​ 2-groups. The maps illustrate​​​‌ spatial patterns of signal‌ intensities across the head,‌​‌ with stronger EEG desynchronisation​​ over the left motor​​​‌ cortex for users with‌ higher performance levels.

8.5.1 Towards neuroadaptive‌ art presentation

Participants: Marc‌​‌ Welter, Axel Bouneau​​​‌, Fabien Lotte.​

External collaborators: Tomas Ward​‌ [DCU, Ireland], Jesus​​ Casal Martínez [UPV, Spain]​​​‌, Jonathan Baum [Inria​ Montpellier], Erin Redmond​‌ [DUC, Ireland].

Both​​ well-being and aesthetic experience​​​‌ are correlated with physiology.​ Thus, optimizing art presentation​‌ to evoke desirable mental​​ states in virtual environments​​​‌ based on physiological states​ could have beneficial effects​‌ on user experience and​​ well-being. However, single trial​​​‌ aesthetic experience decoding from​ physiological signals has not​‌ been well studied. We​​ tested a Support Vector​​​‌ Machine classifier with cardiac​ and electrodermal features to​‌ decode art interest. Although,​​ average performance was poor​​​‌ (54%), the model yielded​ high accuracy beyond chance​‌ level for a few​​ participants. This shows that​​​‌ art interest can, at​ least for some individuals,​‌ be decoded from single​​ trial physiological data 26​​​‌. Beyond this specific​ experiment, we also explored​‌ how BCI-based decoding of​​ aesthetic experience and neuroadaptive​​​‌ art could - in​ theory - be used​‌ for health 27.​​

8.5.2 Performance predictions for​​​‌ running

Participants: Alex Pepi​, Fabien Lotte.​‌

External collaborators: Pierre Gilfriche​​ [Flit Sport], Aurélien​​​‌ Appriou [Flit Sport].​

As part of ERC​‌ Proof-of-Concept (Poc) project SPEARS,​​ in collaboration with the​​​‌ startup FlitSport, we aim​ at better estimating and​‌ predicting runners' running performances​​ and capacity, notably by​​​‌ exploiting their cardiac signals,​ to then be able​‌ to propose them optimal​​ personalized training solution, within​​​‌ FlitSport's FlitCoach app.

When​ runners exercise, their running​‌ capacity can be described​​ with a critical speed​​​‌ ($CS$)​ which is a speed​‌ threshold. In theory, below​​ this speed they can​​​‌ run for hours (indefinitely)​ without being exhausted. Another​‌ parameter describing their capacity​​ is $W^{\text{'}}$,​​​‌ which is the finite​ work capacity available above​‌ $CS$. These​​ parameters can be used​​​‌ to estimate the performance​ of a runner.

They​‌ can be estimated by​​ using only intermittent exercises​​​‌ (alternating runing at near-maximum​ speed and slow running​‌ to rest), because in​​ this kind of exercise,​​​‌ we can see the​ runner's limits. The existing​‌ $CPW$ model​​ is used to predict​​​‌ performances. However, this model​ is used to predict​‌ runner's performance at a​​ specific moment and requires​​​‌ a lot of data​ for estimation (all data​‌ available to be the​​ most accurate).

For the​​​‌ purpose of training personnalization,​ the most relevant estimation​‌ about runners' capacity is​​ the prediction of their​​​‌ future performance, especially with​ a competition coming soon.​‌ There is an existing​​ model that can predict​​​‌ performance in the future​ called the $B\mathrm{a‌}nis\mathrm{t}er$ model. Nonetheless,​​​‌ this model is limited​ because it requires a​‌ lot of data and​​ is not really good​​​‌ at predicting performance in​ a far future (around​‌ 2 or 3 months​​ maximum). Moreover, the parameters​​​‌ are sometimes tricky to​ optimize depending on the​‌ runner's profile.

Consequently, we​​ designed, implemented and validated​​​‌ two new models for​ performance estimation and prediction:​‌

• Riemannian performance estimator:​​ We used Riemannian geometry​​ AI, using the runners'​​​‌ cardiac signals. It uses‌ as input a representation‌​‌ of specific running exercises​​ as covariance matrices, integrating​​​‌ CS and running data.‌ Here, the heart rate‌​‌ (BPM, beat per minutes)​​ was estimated from runners'​​​‌ smart watches or cardiac‌ belts. Using this new‌​‌ approach, the critical speed​​ can be estimated robustly​​​‌ with the data of‌ a short period of‌​‌ exercises only. It proved​​ very robust, reaching a​​​‌ mean squared error (MSE)‌ of around 0.09 only.‌​‌ A significant advantage of​​ this model is the​​​‌ ability to predict performance‌ at a specific moment‌​‌ from only one intermittent​​ exercise (i.e., from little​​​‌ training data, which previoulsy‌ proposed models cannot do).‌​‌ Another key point was​​ the availability of labelled​​​‌ training data provided by‌ Flit Coach. This model‌​‌ was also tested on​​ totally new runners' profiles​​​‌ on Flit Coach's application,‌ and was really accurate‌​‌ according to an evaluation​​ by the runners.
• Long-term​​​‌ performance prediction: We‌ developed a new approach‌​‌ using AI able to​​ take as input the​​​‌ volume of training for‌ the last month and‌​‌ the initial estimated level,​​ and that can predict​​​‌ the critical speed (=‌ performance) for the next‌​‌ month. In fact, if​​ the runner knows his​​​‌ training volume each month,‌ the performance can be‌​‌ predicted in a far​​ future. This model uses​​​‌ a performance by month‌ established by the previous‌​‌ Riemannian algorithm. The final​​ version of this algorithm​​​‌ proved accurate, with a‌ mean squared error around‌​‌ 0.05 only (versus 0.3​​ for the existing Banister​​​‌ model). Moreover, contrary to‌ the Banister model, this‌​‌ new model does not​​ require to be trained​​​‌ and fit on each‌ runner's data, making it‌​‌ more general and applicable​​ in practice, and requiring​​​‌ less training data.

8.5.3‌ Performance predictions for cognitive‌​‌ tasks

Participants: Alex Pepi​​, Fabien Lotte.​​​‌

Running performance could be‌ predicted accurately using Riemannian‌​‌ Geometry algorithms, really popular​​ in EEG classification. For​​​‌ that reason, we explored‌ whether we could also‌​‌ predict cognitive performances with​​ the same approach. To​​​‌ explore this idea, we‌ developped the following protocol:‌​‌

A user is instructed​​ to perform N-back tasks​​​‌ with $N\in \{‌1,2,‌‌3\}$. With​​ an N-back task, letters​​​‌ are displayed on the‌ screen and the user‌​‌ has to determine if​​ the current letter is​​​‌ the same as the‌ previous one (N=1), the‌​‌ one before the previous​​ one (N=2) or the​​​‌ letter displayed three letters‌ before (N=3). The user‌​‌ is wearing an EEG​​ cap to record his/her​​​‌ brain activity and cardiac‌ belt (Polar H9) to‌​‌ record heart rate activity​​ (beats per minutes). Then,​​​‌ different models (Riemannian Suport‌ Vector Classifier (SVC), Tangent‌​‌ Space Classifier (TSC) or​​ Minimum Distance to Mean​​​‌ (MDM)) are used and‌ compared to predict the‌​‌ cognitive performance of the​​ user on this specific​​​‌ task. This model uses‌ in input the covariance‌​‌ matrix between EEG and​​ BPM recorded between the​​​‌ display of two letters.‌ The performance predicted is‌​‌ correct vs incorrect answers​​​‌ fromthe user. The model​ is trained on the​‌ previous user's answers. So​​ far, the experiment was​​​‌ performed only with one​ participant and with 3​‌ trials for each N-back​​ task with 20 letters​​​‌ for each trial. Different​ frequency bands were tested​‌ to find the best​​ accuracy, the beta band​​​‌ was the most efficient,​ and the BPM was​‌ improving accuracy as well.​​

The MDM was the​​​‌ best classifier with an​ accuracy of 76 %,​‌ against 70% for SVC​​ and 68 % for​​​‌ TSC. In the future,​ it would be interesting​‌ to perform such experiments​​ on more subjects, and​​​‌ try different strategies like​ Leave-One-Subject-Out with the different​‌ models. It would also​​ be interesting to try​​​‌ with real ECG recorded​ instead of cardiac belt​‌ recording BPM.

8.5.4 Benchmarking​​ one-class Riemannian EEG classifiers​​​‌ to detect wakefulness under​ general anesthesia

Participants: Valérie​‌ Marissens Cueva, Sébastien​​ Rimbert, Fabien Lotte​​​‌.

External collaborators: Ana​ Maria Cebolla Alvarez [Université​‌ Libre de Bruxelles (ULB),​​ LNMB], Guy Cheron​​​‌ [ULB, LNMB], Claude​ Meistelman [Université de Lorraine]​‌, Seyed Javad Bidgoli​​ [CHU Brugmann], Laurent​​​‌ Bougrain [LORIA, Paris Brain​ Institute].

Current brain​‌ monitors for detecting Accidental​​ Awareness during General Anesthesia​​​‌ (AAGA) remain debated, as​ robust evidence supporting their​‌ effectiveness in reducing AAGA's​​ incidence is lacking. To​​​‌ address this, we propose​ a new brain-computer interface​‌ based on Median Nerve​​ Stimulation (MNS), a painless​​​‌ stimulation that elicits motor​ patterns, to monitor depth​‌ of anesthesia. Specifically, we​​ train our algorithm with​​​‌ post-MNS EEG patterns recorded​ while patients are awake,​‌ enabling the detection of​​ the return to an​​​‌ arousal state during the​ surgery under anesthesia. Since​‌ the anesthesia data is​​ unavailable pre-surgery for BCI​​​‌ calibration, we focus on​ One-Class (OC) approaches. In​‌ this study, we evaluate​​ three OC Riemannian methods​​​‌ for this task: K-Means​ (OC-RKM), Minimum Distance to​‌ the Mean (OC-RMDM) and​​ Support Vector Machine (OC-RSVM).​​​‌ Results indicate that both​ OC-RKM and OCRMDM effectively​‌ delimit an awake state​​ in most subjects, though​​​‌ not all. In contrast,​ OC-RSVM has a lower​‌ performance, possibly due to​​ the use of a​​​‌ Riemannian kernel reference point​ Cref computed as the​‌ mean covariance matrix of​​ the awake class, which​​​‌ may inadequately capture the​ geometry of both classes.​‌ Additionally, ν was not​​ optimized, as its tuning​​​‌ in a one-class context​ remains challenging. Future work​‌ will assess performance regarding​​ the number of electrodes,​​​‌ alternative Cref, and comparisons​ with other one-class algorithms.​‌ An ensemble method will​​ also be considered to​​​‌ improve robustness for depth​ of anesthesia estimation across​‌ subjects, leveraging the strengths​​ of each model. 35​​​‌.

8.5.5 Riemannian fusions​ of EEG-based features for​‌ motor imagery detection under​​ propofol sedation

Participants: Valérie​​​‌ Marissens Cueva, Fabien​ Lotte, Sébastien Rimbert​‌.

External collaborators: Laurent​​ Bougrain [LORIA, Paris Brain​​​‌ Institute], Marie-Constance Corsi​ [Paris Brain Institute],​‌ Camilla Mannino [Paris Brain​​ Institute].

The brain​​​‌ is a complex system​ requiring multimodal approaches to​‌ better understand cognitive or​​ motor functions. Thus, different​​ and complementary electroencephalographic (EEG)​​​‌ neurophysiological features are available‌ at various spatial, frequency,‌​‌ and temporal scales, e.g.,​​ brain connectivity, complexity, or​​​‌ entropy. However, they are‌ usually not investigated all‌​‌ together. In this study,​​ we combine and compare​​​‌ five EEG-based connectivity features‌ with covariance matrices, defining‌​‌ five Riemannian fusion methods​​ and three Euclidean ones​​​‌ as references. We do‌ so for classifying motor‌​‌ imagery EEG signals, both​​ in awake and sedated​​​‌ subjects, with the future‌ goal of detecting accidental‌​‌ awareness during general anesthesia.​​ Covariance matrices alone yielded​​​‌ the best accuracy, with‌ and without sedation. Phase-based‌​‌ connectivity estimators appear to​​ be the most promising​​​‌ fusion with covariances. No‌ significant differences were found‌​‌ between the best fusion​​ of features and that​​​‌ of classifiers. 22.‌

9.1.1 Visits of international​​​‌ scientists

9.1.2 Visits to international‌ teams

9.2.1 Horizon Europe

ANR PROTEUS

Participants:​ Fabien Lotte, Sebastien​‌ Rimbert, Pauline Dreyer​​, David Trocellier.​​

• Title:
  PROTEUS: Measuring, understanding​​​‌ and tackling variabilities in‌ Brain-Computer Interfacing
• Duration:
  2023-2027‌​‌ (3.5 years)
• Partners:
  • Inria​​ Center at the University​​​‌ of Bordeaux, Talence (lead:‌ Fabien Lotte)
  • LAMSADE, Paris‌​‌
  • ISAE-SupAero, Toulouse (lead: Raphaëlle​​ Roy)
  • INSA Rouen, Rouen​​​‌ (lead: Florian Yger)
  • Wisear,‌ Paris (lead: Alain Sirois)‌​‌
• Coordinator:
  Fabien Lotte
• Inria​​ contact:
  Fabien Lotte
• Summary:​​​‌
  Whereas BCI are very‌ promising for various applications‌​‌ they are not reliable.​​ Their reliability degrades even​​​‌ more when used across‌ contexts (e.g., across days,‌​‌ for changing users' states​​ or applications used) due​​​‌ to various sources of‌ variabilities. Project PROTEUS proposes‌​‌ to make BCIs robust​​ to such variabilities by​​​‌ 1) Systematically measuring BCI‌ and brain signal variabilities‌​‌ across various contexts while​​ sharing the collected databases;​​​‌ 2) Characterising, understanding and‌ modelling the variability and‌​‌ their sources based on​​ these new databases; and​​​‌ 3) Tackling these variabilities‌ by designing new machine‌​‌ learning algorithms optimally invariant​​ to them according to​​​‌ our models, and using‌ the resulting BCIs for‌​‌ two practical applications affected​​ by variabilities: tetraplegic BCI​​​‌ user training and auditory‌ attention monitoring at home‌​‌ or in flight.
• Website:​​
  PROTEUS

ANR BCI4IA

Participants:​​​‌ Sebastien Rimbert, Fabien‌ Lotte, Valerie Marissens‌​‌.

• Title:
  BCI4IA: a​​ New BCI Paradigm To​​​‌ Detect Intraoperative Awareness During‌ General Anesthesia
• Duration:
  2023-2027‌​‌ (4 years)
• Partners:
  • Inria​​ Center at the University​​​‌ of Bordeaux, Talence (lead:‌ Fabien Lotte)
  • LORIA, Nancy‌​‌ (lead: Laurent Bougrain)
  • CHRU​​ Nancy, Nancy (lead: Claude​​​‌ Meistelman)
  • CHU Brugmann, Brussels‌ (lead: Denis Schwartz)
  • Univ.‌​‌ Libre Bruxelles, Brussels (lead:​​ Anna Cebolla)
• Coordinator:
  Claude​​​‌ Meistelman
• Inria contact:
  Sebastien‌ Rimbert
• Summary:
  The BCI4IA‌​‌ project aims to design​​ a brain-computer interface to​​​‌ enable reliable general anesthesia‌ (GA) monitoring, in particular‌​‌ to detect intraoperative awareness.​​ Currently, there is no​​​‌ satisfactory solution to do‌ so whereas it causes‌​‌ severe post-traumatic stress disorder.​​ "I couldn't breathe, I​​​‌ couldn't move or open‌ my eyes, or even‌​‌ tell the doctors I​​ wasn't asleep." This testimony​​​‌ shows that a patient's‌ first reaction during an‌​‌ intraoperative awareness is usually​​ to move to alert​​​‌ the medical staff. Unfortunately,‌ during most surgery, the‌​‌ patient is curarized, which​​ causes neuromuscular block and​​​‌ prevents any movement. To‌ prevent intraoperative awareness, we‌​‌ propose to study motor​​ brain activity under GA​​​‌ using electroencephalography (EEG) to‌ detect markers of motor‌​‌ intention (MI) combined with​​ general brain markers of​​​‌ consciousness. We will analyze‌ a combination of MI‌​‌ markers (relative powers, connectivity)​​ under the propofol anesthetics,​​​‌ with a brain-computer interface‌ based on median nerve‌​‌ stimulation to amplify them.​​ Doing so will also​​​‌ require to design new‌ machine learning algorithms based‌​‌ on one-class (rest class)​​ EEG classification, since no​​​‌ EEG examples of the‌ patient's MI under GA‌​‌ are available to calibrate​​ the BCI. Our preliminary​​​‌ results are very promising‌ to bring an original‌​‌ solution to this problem​​ which causes serious traumas.​​​‌
• Website:
  BCI4IA

ANR STIM-BCI‌

Participants: Sebastien Rimbert,‌​‌ Fabien Lotte, Loic​​ Bechon.

• Title:
  STIM-BCI:​​​‌ Using Somesthetic Stimulations to‌ Design Next-Generation Motor Brain-Computer‌​‌ Interfaces
• Duration:
  2024–2028 (4​​​‌ years)
• Partners:
  • Inria Center​ at the University of​‌ Bordeaux, Talence (lead: Sébastien​​ Rimbert)
  • University of Lorraine,​​​‌ Nancy (lead: Stéphanie Fleck)​
  • Université Libre de Bruxelles,​‌ Brussels (lead: Ana Maria​​ Cebolla)
  • RIKEN AIP, Tokyo​​​‌ (lead: Tomasz Rutkowski)
• Coordinator:​
  Sebastien Rimbert
• Inria contact:​‌
  Sebastien Rimbert
• Summary:
  The​​ STIM-BCI project aims to​​​‌ develop and formalize a​ new motor brain-computer interface​‌ (BCI) paradigm based on​​ somaesthetic stimulation (SOM-BCI). It​​​‌ builds on experimental evidence​ showing that painless somesthetic​‌ stimulation, in particular median​​ nerve stimulation, enhances motor​​​‌ intention-related EEG patterns and​ reduces their variability. STIM-BCI​‌ focuses on (i) defining​​ and characterizing SOM-BCIs within​​​‌ the BCI taxonomy, (ii)​ evaluating their performance, usability​‌ and acceptability, and (iii)​​ designing dedicated signal processing​​​‌ and machine learning methods​ adapted to their specific​‌ neurophysiological signatures. The project​​ ultimately aims to improve​​​‌ the robustness and usability​ of MI-based BCIs and​‌ to extend their use​​ beyond laboratory settings.
• Website:​​​‌
  STIM-BCI

AEx D-CodeBrain

Participants:​ Sebastien Rimbert, Fabien​‌ Lotte.

• Title:
  D-CodeBrain:​​ Decoding Motor Brain Activity​​​‌ for User Profiling in​ Motor BCIs
• Duration:
  2025–2027​‌ (Action Exploratoire Inria)
• Partners:​​
  • Inria Center at the​​​‌ University of Bordeaux (team​ Potioc)
  • University of Lorraine​‌ (team PErSEUs)
• Coordinator:
  Sebastien​​ Rimbert
• Inria contact:
  Sebastien​​​‌ Rimbert
• Summary:
  The D-CodeBrain​ project aims to design​‌ an exploratory tool for​​ profiling users of motor​​​‌ imagery-based brain-computer interfaces (MI-BCIs)​ by decoding motor-related EEG​‌ activity elicited by a​​ single somaesthetic stimulation. Building​​​‌ on evidence that somesthetic​ stimulations, such as median​‌ nerve stimulation, induce robust​​ and informative motor cortical​​​‌ responses, the project seeks​ to characterize inter-individual variability​‌ in motor brain activity​​ and its relation to​​​‌ user states such as​ fatigue, workload, stress or​‌ motivation. D-CodeBrain combines EEG​​ signal processing, machine learning​​​‌ and user-centered design to​ generate individualized motor and​‌ cognitive profiles, with the​​ objective of improving BCI​​​‌ reliability, usability and adaptability​ beyond laboratory settings.

NEURO-PULSE​‌

Participants: Sebastien Rimbert,​​ Fabien Lotte, Grégoire​​​‌ Cane.

• Title:
  NEURO-PULSE:​ Neuro-Prognostication of Brain-Injured Patients​‌ Using EEG and Artificial​​ Intelligence
• Duration:
  2025–2028
• Partners:​​​‌
  • Inria Center at the​ University of Bordeaux (team​‌ Potioc)
  • CHU Pellegrin, Bordeaux​​ (Neuro-Intensive Care Unit)
• Coordinator:​​​‌
  Sebastien Rimbert
• Inria contact:​
  Sebastien Rimbert
• Summary:
  The​‌ NEURO-PULSE project aims to​​ improve neuro-prognostication in severely​​​‌ brain-injured patients, and in​ particular in coma, by​‌ developing a passive and​​ clinically compatible brain-computer interface​​​‌ (BCI) based on EEG​ analysis and machine learning.​‌ The project focuses on​​ the detection of reliable​​​‌ cerebral markers of consciousness​ and brain engagement, independent​‌ of any behavioral or​​ motor response, addressing situations​​​‌ of cognitive-motor dissociation. NEURO-PULSE​ combines advanced EEG signal​‌ processing, machine learning methods​​ adapted to noisy and​​​‌ heterogeneous clinical data, and​ close collaboration with neuro-intensive​‌ care clinicians, with the​​ objective of supporting medical​​​‌ decision-making and facilitating the​ clinical transfer of BCI​‌ technologies in real hospital​​ environments.

10.1.1​ Scientific events: organisation

10.1.2 Scientific events: selection‌

10.1.3 Journal

10.1.4 Invited‌​‌ talks

• "Aesthetic Experience Decoding​​ with multi-modal passive Brain-Computer-Interfaces",​​​‌ Department of Computer Science,‌ VU Amsterdam, Netherlands, February‌​‌ 2025, Marc Welter
• "The​​ promises and pitfalls of​​​‌ passive Neuroart Brain-Computer-Interfaces for‌ health and well-being", Third‌​‌ International Workshop on Complex​​ Systems Science and Health​​​‌ Neuroscience, Torun, Poland, June‌ 2025. Marc Welter
• "Decoding‌​‌ Aesthetic Experience with Brain-Computer-Interfaces",​​ Department of Behavioral and​​​‌ Movement Sciences, VU Amsterdam,‌ Netherlands, November 2025. Marc‌​‌ Welter
• "Artificial Intelligence for​​ Neurotechnologies: opportunities and biases",​​​‌ USERN Congress 2025 -‌ virtual congress invited keynote,‌​‌ online, November 2025, Fabien​​ Lotte
• "Brain-Computer Interaction design:​​​‌ from basic science to‌ applications", Queen's University Belfast,‌​‌ Belfast, UK, invited talk,​​ May 2025, Fabien Lotte​​​‌
• "Machine learning for EEG-BCI:‌ recent Riemannian advances and‌​‌ guidelines to avoid biases",​​ BCI and Neurotechnology Spring​​​‌ School, online, Keynote speaker,‌ May 2025, Fabien Lotte‌​‌
• "Machine Learning for EEG-based​​ BCIs: recent advances and​​​‌ ambushes", Institute of Global‌ Innovation Research Open Seminar,‌​‌ Tokyo University of Agriculture​​​‌ and Technology, Tokyo, Japan,​ invited talk, March 2025,​‌ Fabien Lotte
• “Using Somesthetic​​ Stimulation to Improve BCI​​​‌ Performance”, invited talk, Institute​ of Global Innovation Research​‌ Open Seminar, Tokyo University​​ of Agriculture and Technology,​​​‌ Tokyo, Japan, March 2025,​ Sebastien Rimbert
• "Overcoming subject​‌ specific variability for MI-BCI​​ with machine learning and​​​‌ deep learning", BTU (Branderburg​ Technological University), Neuroadaptive Human​‌ computer Interface lab, Germany,​​ April 2025, David Trocellier​​​‌

10.1.5 Leadership within the​ scientific community

• Member of​‌ the board of CORTICO,​​ the French BCI research​​​‌ society (Sebastien Rimbert​ , Fabien Lotte )​‌

10.1.6 Scientific expertise

Member​​ of Scientific Advisory Boards:​​​‌

• Member of the Scientific​ Advisory Board of the​‌ Research Training Group “Neuromodulation”,​​ University of Oldenburg, Germany,​​​‌ (Fabien Lotte )​

Evaluation of projects for​‌ the following research grants/bodies:​​

• SFRMS (Fabien Lotte​​​‌ )
• MIAI (Fabien​ Lotte )

10.2.1 Teaching

• Master:​ Neuroergonomie, 6h eqTD, M2​‌ Cognitive science, Université de​​ Bordeaux (Fabien Lotte​​​‌ )
• Master: Introduction à​ la Neuroergonomie, 3h eqTD,​‌ M1 Cognitive science, Université​​ de Bordeaux (Fabien​​​‌ Lotte )
• Master: Interface​ cerveau-ordinateur, 35h eqTD, M2​‌ Cognitive science, Université de​​ Lorraine (Sébastien Rimbert​​​‌ )
• Master: Brain-Computer Interfaces,​ 17h eqTD, M2 Cognitive​‌ science IDMC, Université de​​ Lorraine (Valérie Marissens​​​‌ Cueva )
• Master: Scientific​ Protocols, 5h TD +​‌ 5h CM, M2 Cognitive​​ science IDMC, Université de​​​‌ Lorraine (Valérie Marissens​ Cueva )
• Engineering school:​‌ Immersion and interaction with​​ visual worlds, 4,5h eqtd,​​​‌ Graduate Degree Artificial Intelligence​ and Advanced Visual Computing,​‌ Ecole Polytechnique Palaiseau (​​Fabien Lotte )
• Engineering​​​‌ school: Artificial Intelligence, 32h​ eqTD, 2nd year, Telecom​‌ Nancy (Valérie Marissens​​ Cueva )
• Bachelor: Travaux​​​‌ encadrée de recherche, 20h​ Culture et Compétences Numériques​‌ - PIX, L2,Université de​​ Bordeaux (Pauline Dreyer​​​‌ )
• Bachelor: Programmation et​ applications interactives, 32h eqTD,​‌ L1, Université de Bordeaux​​ (David Trocellier )​​​‌
• Bachelor: Travaux encadrée de​ recherche, 5h eqTD, L3,​‌ Université de Bordeaux (​​David Trocellier )
• IFMK:​​​‌ Jury of English evaluation,​ 7h eqTD, K4 IFMK,​‌ IFMK du CHU De​​ Bordeaux (Pauline Dreyer​​​‌ )

10.2.2 Supervision

• PhD​ thesis: David Trocellier (​‌Fabien Lotte 50%)
• PhD​​ thesis: Marc Welter (​​​‌Fabien Lotte 100%)
• PhD​ thesis: Côme Annicchiarico (​‌Fabien Lotte 25%)
• PhD​​ thesis: Pauline Dreyer (​​​‌Fabien Lotte 50%)
• PhD​ thesis: Valerie Marissens-Cueva (​‌Fabien Lotte 33%, Sebastien​​ Rimbert 33%)
• PhD thesis:​​​‌ Thibault de Surrel (​Fabien Lotte 15%)
• PhD​‌ thesis: Loic Bechon (​​Sebastien Rimbert 33%, Fabien​​​‌ Lotte 33%)
• PhD thesis:​ Juliette Meunier (Fabien​‌ Lotte 33%, Sebastien Rimbert​​ 33%)
• PhD thesis: Manon​​​‌ Bourdil (Fabien Lotte​ 50%)
• Master internship: Camille​‌ Cousin (Fabien Lotte​​ 50%)
• Master internship: Manon​​​‌ Bourdil (Pauline Dreyer​ 100%)

10.2.3 Juries

PhD​‌ committees:

• Hannah Pulferer, TU​​ Graz, Austria (Reviewer) (​​​‌Fabien Lotte )
• Felix​ Schroeder, LMU, UK (Reviewer)​‌ (Fabien Lotte )​​
• Gabriel Wagner vom Berg,​​​‌ TU Berlin, Germany (Reviewer)​ (Fabien Lotte )​‌
• Evy Van Weelden, Univ.​​ Tilburg, The Netherlands (Reviewer)​​ (Fabien Lotte )​​​‌
• Nicolas Ivanov, Univ. Toronto,‌ Canada (Reviewer) (Fabien‌​‌ Lotte )
• Imen Ayadi,​​ Centrale Paris, France (Examiner)​​​‌ (Fabien Lotte )‌
• Mathias Rihet, ISAE Supaéro,‌​‌ France (Examiner) (Fabien​​ Lotte )
• Yassine El-Houaidi,​​​‌ IMT Brest, France (Guest)‌ (Fabien Lotte )‌​‌
• Claire Dussard, ICM Institute​​ Paris (Sebastien Rimbert​​​‌ )

Habilitation (HDR) committee:‌

• Camille Jeunet-Kelway, Univ. Bordeaux‌​‌ / CNRS, France (Guest)​​ (Fabien Lotte )​​​‌

PhD follow-up committees (CSI):‌

• Camilla Mannino, Sorbonne Université‌​‌ / Inria (Fabien​​ Lotte )
• Jarod Levy,​​​‌ Univ. Paris-Saclay / Meta‌ (Fabien Lotte )‌​‌
• Maeva Andriantsoamberomanga, Univ. Bordeaux​​ / Inria (Fabien​​​‌ Lotte )

10.3.1 Specific official responsibilities‌​‌ in science outreach structures​​

• Vice President of Events​​​‌ and member of the‌ Communications team at Ascoergo,‌​‌ an association dedicated to​​ promoting cognitive science by​​​‌ organizing monthly outreach events‌ where researchers present their‌​‌ work to the general​​ public (Juliette Meunier​​​‌ , Loïc Bechon ).‌

10.3.2 Productions (articles, videos,‌​‌ podcasts, serious games, ...)​​

• "Et si demain l'IA​​​‌ lisait dans nos pensées‌ ?", podcast Déssamblons le‌​‌ numérique, épisode 14, septembre​​ 2025 (Fabien Lotte​​​‌ )
• Participation in a‌ podcast recorded by the‌​‌ publisher of the Neural​​ Interfaces book, aimed at​​​‌ promoting its content (‌Pauline Dreyer , David‌​‌ Trocellier , Fabien Lotte​​ )
• Participation in the​​​‌ recording of an episode‌ of L'Esprit Sorcier,‌​‌ recorded in Paris, with​​ the goal of explaining​​​‌ the principles, challenges, and‌ ongoing research at Inria‌​‌ in Brain Computer Interfaces​​ to the general public.​​​‌ (Fabien Lotte ,‌ Pauline Dreyer )
• Fondation‌​‌ pour la recherche médicale​​ : Accident vasculaires cérébraux​​​‌ : une interface cerveau‌ machine pour améliorer la‌​‌ rééducation (article)​​ (David Trocellier )​​​‌
• Fondation pour la recherche‌ médicale : Accident vasculaires‌​‌ cérébraux : une interface​​ cerveau machine pour améliorer​​​‌ la rééducation (vidéo‌) (David Trocellier‌​‌ , Fabien Lotte )​​

10.3.3 Participation in Live​​​‌ events

• Presenting our science‌ outreach tool which is‌​‌ called BrainKart 37,​​ in order to show​​​‌ a tangible demonstration of‌ what can be done‌​‌ in the neurotechnology field​​ at the Handitech forum,​​​‌ Paris, France (David‌ Trocellier , Juliette Meunier‌​‌ )
• Presenting at Viva​​ Technology, with our German​​​‌ collaborator from DFKI, the‌ work we are doing‌​‌ on the NEARBY project,​​ as part of the​​​‌ French-German Tech Lab (‌Juliette Meunier )
• Presenting‌​‌ Brain Kart, a kart​​ controlled by brain activity​​​‌ at the "Forum des‌ sciences cognitives", in Paris‌​‌ (David Trocellier ,​​ Juliette Meunier , Loïc​​​‌ Bechon )

10.3.4 Others‌ science outreach relevant activities‌​‌

• Presentation of our team​​ and our science outreach​​​‌ tool Brain kart to‌ students from junior high‌​‌ school (3ème), highschool and​​ first year of bachelor​​​‌ degree (Fabien Lotte‌ , Juliette Meunier ,‌​‌ Loïc Bechon , Pauline​​ Dreyer , Alex Pepi​​​‌ , Camille Cousin ,‌ Manon Bourdil ).

International journals

• 13‌ articleD. E.David‌​‌ E Carlson, R.​​Ricardo Chavarriaga, Y.​​​‌Yiling Liu, F.‌Fabien Lotte and B.-L.‌​‌Bao-Liang Lu. The​​ NERVE-ML (neural engineering reproducibility​​​‌ and validity essentials for‌ machine learning) checklist: ensuring‌​‌ machine learning advances neural​​ engineering *.Journal​​​‌ of Neural Engineering22‌2March 2025,‌​‌ 021002HALDOIback​​ to text
• 14 article​​​‌S.Stephanie Cernera,‌ T.Tan Gemicioglu,‌​‌ J.Julia Berezutskaya,​​ R.Richard Csaky,​​​‌ M.Maxime Verwoert,‌ D.Daniel Polyakov,‌​‌ S.Sotirios Papadopoulos,​​ V.Valeria Spagnolo,​​​‌ J. G.Juliana Gonzalez‌ Astudillo, S.Satyam‌​‌ Kumar, H.Hussein​​ Alawieh, D.Dion​​​‌ Kelly, J. R.‌Joanna R G Keough‌​‌, A.Araz Minhas​​, M.Matthias Dold​​​‌, Y.Yiyuan Han‌, A.Alexander Mcclanahan‌​‌, M.Mousa Mustafa​​, J. J.Juan​​​‌ Jose Gonzalez-Espana, F.‌Florencia Garro, A.‌​‌Angela Vujic, K.​​Kriti Kacker, C.​​​‌Christoph Kapeller, S.‌Simon Geukes, C.‌​‌Ceci Verbaarschot, M.​​Michael Wimmer, M.​​​‌Mushfika Sultana, S.‌Sara Ahmadi, C.‌​‌Christian Herff, A.​​ I.Andreea Ioana Sburlea​​​‌, C.Camille Jeunet‌, D. E.David‌​‌ E Thompson, M.​​Marianna Semprini, R.​​​‌Richard Andersen, S.‌Sergey Stavisky, E.‌​‌Eli Kinney-Lang, F.​​Fabien Lotte, J.​​​‌Jordy Thielen, X.‌Xing Chen, V.‌​‌Victoria Peterson, A.​​Aysegul Gunduz, T.​​​‌Theresa Vaughan and D.‌Davide Valeriani. Master‌​‌ classes of the tenth​​ international brain–computer interface meeting:​​​‌ showcasing the research of‌ BCI trainees.Journal‌​‌ of Neural Engineering22​​2February 2025,​​​‌ 022001HALDOIback‌ to text
• 15 article‌​‌T.Tommaso Dorigo,​​ G. D.Gary D​​​‌ Brown, C.Carlo‌ Casonato, A.Artemi‌​‌ Cerdà, J.Joseph​​ Ciarrochi, M. D.​​​‌Mauro Da Lio,‌ N.Nicole D’souza,‌​‌ N. R.Nicolas R​​ Gauger, S. C.​​​‌Steven C Hayes,‌ S. G.Stefan G‌​‌ Hofmann, R.Robert​​ Johansson, M.Marcus​​​‌ Liwicki, F.Fabien‌ Lotte, J. J.‌​‌Juan J Nieto,​​ G.Giulia Olivato,​​​‌ P.Peter Parnes,‌ G.George Perry,‌​‌ A.Alice Plebe,​​ I. M.Idupulapati M​​​‌ Rao, N.Nima‌ Rezaei, F.Fredrik‌​‌ Sandin, A.Andrey​​​‌ Ustyuzhanin, G.Giorgio​ Vallortigara, P.Pietro​‌ Vischia and N.Niloufar​​ Yazdanpanah. Artificial Intelligence​​​‌ in Science and Society:​ the Vision of USERN​‌.IEEE Access13​​2025, 15993-16054HAL​​​‌DOIback to text​
• 16 articleV.Valérie​‌ Marissens Cueva, L.​​Laurent Bougrain, F.​​​‌Fabien Lotte and S.​Sébastien Rimbert. Reliable​‌ predictor of BCI motor​​ imagery performance using median​​​‌ nerve stimulation.Journal​ of Neural EngineeringMarch​‌ 2025HALDOIback​​ to text
• 17 article​​​‌A.Aline Roc,​ L.Léna Kolodzienski,​‌ P.Pauline Dreyer,​​ A.Aurélien Appriou,​​​‌ T.Thibaut Monseigne and​ F.Fabien Lotte.​‌ Evolution of users’ subjective​​ experience over three training​​​‌ sessions with an EEG​ Motor-Imagery Brain–Computer Interface (MI-BCI)​‌.Brain Research1872​​2025, 150085HAL​​​‌DOIback to text​
• 18 articleT.Thibault​‌ de Surrel, S.​​Sylvain Chevallier, F.​​​‌Fabien Lotte and F.​Florian Yger. Geometry-Aware​‌ visualization of high dimensional​​ Symmetric Positive Definite matrices​​​‌.Transactions on Machine​ Learning Research JournalFebruary​‌ 2025HALback to​​ text

International peer-reviewed conferences​​​‌

• 19 inproceedingsL.Loïc​ Bechon, F.Fabien​‌ Lotte, S.Stéphanie​​ Fleck and S.Sébastien​​​‌ Rimbert. Optimizing Median​ Nerve Stimulation-based BCI: Impact​‌ of Timing and Intensity​​ on ERD/ERS.Journées​​​‌ CORTICO 2025 - COllectif​ pour la Recherche Transdisciplinaire​‌ sur les Interfaces Cerveau-Ordinateur​​Lyon, FranceMay 2025​​​‌HALback to text​
• 20 inproceedingsV. M.​‌Valérie Marissens Cueva,​​ F.Fabien Lotte,​​​‌ L.Laurent Bougrain and​ S.Sébastien Rimbert.​‌ From Post-Median Nerve Stimulation​​ ERD to MI-BCI Expertise​​​‌ Prediction.BCI 2025​ - 13th International Winter​‌ Conference on Brain-Computer Interface​​Seoul, South KoreaIEEE​​​‌February 2025, 1-6​HALDOIback to​‌ text
• 21 inproceedingsV.​​Valérie Marissens Cueva,​​​‌ F.Fabien Lotte,​ L.Laurent Bougrain and​‌ S.Sébastien Rimbert.​​ Quantifying Inter-and Intra-Subject Variability​​​‌ of Sensorimotor Desynchronization Induced​ by Median Nerve Stimulation​‌ and Motor Imagery for​​ BCI.EMBS 2025​​​‌ - 47th Annual International​ Conference of the IEEE​‌ Engineering in Medicine and​​ Biology SocietyCopenhague, Denmark​​​‌July 2025HALback​ to text
• 22 inproceedings​‌V.Valérie Marissens Cueva​​, C.Camilla Mannino​​​‌, M.-C.Marie-Constance Corsi​, F.Fabien Lotte​‌, S.Sébastien Rimbert​​ and L.Laurent Bougrain​​​‌. Riemannian fusions of​ EEG-based features for motor​‌ imagery detection under propofol​​ sedation.MLSP 2025​​​‌ - IEEE International Workshop​ on Machine Learning for​‌ Signal ProcessingIstanbul, Turkey​​August 2025HALback​​​‌ to text
• 23 inproceedings​T.Thibault de Surrel​‌, F.Fabien Lotte​​, S.Sylvain Chevallier​​​‌ and F.Florian Yger​. Wrapped Gaussian on​‌ the manifold of Symmetric​​ Positive Definite Matrices.​​​‌Proceedings of the 42​ nd International Conference on​‌ Machine Learning, Vancouver, Canada.​​ PMLR 267, 2025.ICML​​​‌ 2025 - 42nd International​ Conference on Machine Learning​‌Vancouver, CanadaJuly 2025​​HALback to text​​​‌back to text
• 24​ inproceedingsT.Thibault de​‌ Surrel, F.Florian​​ Yger, F.Fabien​​ Lotte and S.Sylvain​​​‌ Chevallier. A probabilistic‌ view on Riemannian machine‌​‌ learning models for SPD​​ matrices.Lecture Notes​​​‌ in Computer Science, vol‌ 16035GSI 2025 -‌​‌ 7th International Conference on​​ Geometric Science of Information​​​‌Saint Malo, FranceOctober‌ 2025, 142–151HAL‌​‌DOIback to text​​back to text
• 25​​​‌ inproceedingsD.David Trocellier‌, B.Bernard N'Kaoua‌​‌ and F.Fabien Lotte​​. BCI classifiers integrating​​​‌ measures of variability factors:‌ a mini-review.Proceedings‌​‌ of the NeuroAdaptive Technology​​ Conference 2025NeuroAdaptive Technology​​​‌ Conference 2025 (NAT'25)Berlin,‌ GermanyApril 2025HAL‌​‌back to text
• 26​​ inproceedingsM.Marc Welter​​​‌, A.Axel Bouneau‌, M.Martinez Jesus‌​‌ Casal, R.Redmond​​ Erin, J.Jonathan​​​‌ Baum, T.Tomás‌ Ward and F.Fabien‌​‌ Lotte. Physiological Single​​ trial decoding of art​​​‌ interest.Proceedings of‌ NeuroAdaptive Technolgy Conference 2025‌​‌NAT 2025 - NeuroAdaptive​​ Technolgy ConferenceBerlin, Germany​​​‌April 2025HALback‌ to text
• 27 inproceedings‌​‌M.Marc Welter,​​ T. M.Tomasz Maciej​​​‌ Rutkowski and F.Fabien‌ Lotte. Passive Neuroart‌​‌ BCI for Health: A​​ Perspective.Proceedings of​​​‌ the 11th International BCI‌ MeetingBCI 2025 -‌​‌ 11th International BCI Meeting​​Banff, CanadaJune 2025​​​‌HALback to text‌
• 28 inproceedingsM.Mathilde‌​‌ Yousefi, J.Jérémy​​ Frey, S.Sébastien​​​‌ Rimbert, A. G.‌Altamira Gabriela Herrera and‌​‌ S.Stéphanie Fleck.​​ Sense-It: Towards new types​​​‌ of BCI-based sensorimotor neurofeedback‌ for motor rehabilitation.‌​‌SMC 2025 - IEEE​​ International Conference on Systems,​​​‌ Man, and CyberneticsVienna,‌ AustriaOctober 2025HAL‌​‌back to text

Conferences​​ without proceedings

• 29 inproceedings​​​‌P.Pauline Dreyer,‌ M.Manon Bourdil,‌​‌ R. N.Raphaëlle N​​ Roy and F.Fabien​​​‌ Lotte. Toward understanding‌ intra-user variability in BCI‌​‌ : a multi-session, multi-context​​ experimental protocol.CORTICO​​​‌ 2025 - Journées COllectif‌ pour la Recherche Transdisciplinaire‌​‌ sur les Interfaces Cerveau-Ordinateur​​Lyon, FranceMay 2025​​​‌HALback to text‌
• 30 inproceedingsV.Valérie‌​‌ Marissens Cueva, L.​​Laurent Bougrain, F.​​​‌Fabien Lotte and S.‌Sébastien Rimbert. Median‌​‌ nerve stimulation to predict​​ MI-BCI performances.Journées​​​‌ CORTICO 2025 - COllectif‌ pour la Recherche Transdisciplinaire‌​‌ sur les Interfaces Cerveau-Ordinateur​​Lyon, FranceMay 2025​​​‌HALback to text‌back to text

Scientific‌​‌ book chapters

• 31 inbook​​F.Fabien Lotte,​​​‌ P.Pauline Dreyer,‌ S.Sébastien Rimbert,‌​‌ D.David Trocellier and​​ M.Marc Welter.​​​‌ How to build a‌ brain–computer interface from beginning‌​‌ to end.Neural​​ InterfacesElsevier2025,​​​‌ 115-141HALDOIback‌ to text

Reports &‌​‌ preprints

• 32 miscC.​​Côme Annicchiarico, F.​​​‌Fabien Lotte and J.‌Jérémie Mattout. An‌​‌ Active Inference perspective on​​ Neurofeedback Training.2025​​​‌HALback to text‌
• 33 miscS.Simon‌​‌ Kojima and F.Fabien​​ Lotte. New Metrics​​​‌ of Event-Related (De)Synchronization Temporal‌ Variability Explain Motor Imagery-based‌​‌ BCI Performance.January​​ 2026HALback to​​​‌ text

Other scientific publications‌

• 34 inproceedingsP.Pauline‌​‌ Dreyer, R. N.​​​‌Raphaëlle N Roy and​ F.Fabien Lotte.​‌ Building a taxonomy of​​ variability factors in active​​​‌ BCI.BCI 2025​ - 11th International BCI​‌ Meeting Building Momentum: Fostering​​ Collaboration in BCIBanff,​​​‌ CanadaJune 2025HAL​back to text
• 35​‌ inproceedingsV.Valérie Marissens​​ Cueva, S.Sébastien​​​‌ Rimbert, A. M.​Ana Maria Cebolla Alvarez​‌, G.Guy Cheron​​, C.Claude Meistelman​​​‌, S. J.Seyed​ Javad Bidgoli, F.​‌Fabien Lotte and L.​​Laurent Bougrain. Benchmarking​​​‌ one-class Riemannian EEG classifiers​ to detect wakefulness under​‌ general anesthesia.BCI​​ 2025 - 11th International​​​‌ BCI MeetingBanff, Canada​June 2025HALback​‌ to text
• 36 inproceedings​​J.Juliette Meunier,​​​‌ S.Sébastien Rimbert,​ A.Arnaud Prouzeau and​‌ F.Fabien Lotte.​​ Towards a protocol to​​​‌ find neurophysiological markers of​ collaborations.CORTICO 2025​‌ - COllectif pour la​​ Recherche Transdisciplinaire sur les​​​‌ Interfaces Cerveau-OrdinateurLyon, France​May 2025HAL
• 37​‌ inproceedingsD.David Trocellier​​, J.Juliette Meunier​​​‌, L.Loïc Bechon​, P.Philomène Motti​‌, B.Bernard N'Kaoua​​ and F.Fabien Lotte​​​‌. BRAINKART : A​ tangible open-source fast calibration​‌ BCI for science popularisation​​.CORTICO 2025 -​​​‌ Journées pour le COllectif​ pour la Recherche Transdisciplinaire​‌ sur les Interfaces Cerveau-Ordinateur​​Lyon, FranceMay 2025​​​‌HALback to text​