Philipp Thölke, Gianni De Fabritiis
International Conference on Learning Representations (ICLR) | 2022
The prediction of quantum mechanical properties is historically plagued by a trade-off between accuracy and speed. Machine learning potentials have previously shown great success in this domain, reaching increasingly better accuracy while maintaining computational efficiency comparable with classical force fields. In this work we propose TorchMD-NET, a novel equivariant transformer (ET) architecture, outperforming state-of-the-art on MD17, ANI-1, and many QM9 targets in both accuracy and computational efficiency. Through an extensive attention weight analysis, we gain valuable insights into the black box predictor and show differences in the learned representation of conformers versus conformations sampled from molecular dynamics or normal modes. Furthermore, we highlight the importance of datasets including off-equilibrium conformations for the evaluation of molecular potentials.
Philipp Thölke, Yorguin Jose Mantilla Ramos, Hamza Abdelhedi, Charlotte Maschke, Arthur Dehgan, Yann Harel, Anirudha Kemtur, Loubna Mekki Berrada, Myriam Sahraoui, Tammy Young, Vanessa Hadid, Etienne Combrissone, Jordan O’Byrne, Karim Jerbi
NeuroImage | 2023
Machine learning (ML) is increasingly used in cognitive, computational and clinical neuroscience. The reliable and efficient application of ML requires a sound understanding of its subtleties and limitations. Training ML models on datasets with imbalanced classes is a particularly common problem, and it can have severe consequences if not adequately addressed. With the neuroscience ML user in mind, this paper provides a didactic assessment of the class imbalance problem and illustrates its impact through systematic manipulation of data imbalance ratios in (i) simulated data and (ii) brain data recorded with electroencephalography (EEG), magnetoencephalography (MEG) and functional magnetic resonance imaging (fMRI). Our results illustrate how the widely-used Accuracy (Acc) metric, which measures the overall proportion of successful predictions, yields misleadingly high performances, as class imbalance …
Wenhui Cui, Woojae Jeong, Philipp Thölke, Takfarinas Medani, Karim Jerbi, Anand A Joshi, Richard M Leahy
IEEE International Symposium on Biomedical Imaging | 2024
To handle the scarcity and heterogeneity of electroencephalography (EEG) data for Brain-Computer Interface (BCI) tasks, and to harness the power of large publicly available data sets, we propose Neuro-GPT, a foundation model consisting of an EEG encoder and a GPT model. The foundation model is pre-trained on a large-scale data set using a self-supervised task that learns how to reconstruct masked EEG segments. We then fine-tune the model on a Motor Imagery Classification task to validate its performance in a low-data regime (9 subjects). Our experiments demonstrate that applying a foundation model can significantly improve classification performance compared to a model trained from scratch, which provides evidence for the generalizability of the foundation model and its ability to address challenges of data scarcity and heterogeneity in EEG.
Maciej Majewski, Adrià Pérez, Philipp Thölke, Stefan Doerr, Nicholas E Charron, Toni Giorgino, Brooke E Husic, Cecilia Clementi, Frank Noé, Gianni De Fabritiis
Nature communications | 2023
A generalized understanding of protein dynamics is an unsolved scientific problem, the solution of which is critical to the interpretation of the structure-function relationships that govern essential biological processes. Here, we approach this problem by constructing coarse-grained molecular potentials based on artificial neural networks and grounded in statistical mechanics. For training, we build a unique dataset of unbiased all-atom molecular dynamics simulations of approximately 9 ms for twelve different proteins with multiple secondary structure arrangements. The coarse-grained models are capable of accelerating the dynamics by more than three orders of magnitude while preserving the thermodynamics of the systems. Coarse-grained simulations identify relevant structural states in the ensemble with comparable energetics to the all-atom systems. Furthermore, we show that a single coarse-grained potential …
Antoine Bellemare-Pepin, François Lespinasse, Philipp Thölke, Yann Harel, Kory Mathewson, Jay A Olson, Yoshua Bengio, Karim Jerbi
Scientific Reports | 2026
The recent surge of Large Language Models (LLMs) has led to claims that they are approaching a level of creativity akin to human capabilities. This idea has sparked a blend of excitement and apprehension. However, a critical piece that has been missing in this discourse is a systematic evaluation of LLMs' semantic diversity, particularly in comparison to human divergent thinking. To bridge this gap, we leverage recent advances in computational creativity to analyze semantic divergence in both state-of-the-art LLMs and a substantial dataset of 100,000 humans. We found evidence that LLMs can surpass average human performance on the Divergent Association Task, and approach human creative writing abilities, though they fall short of the typical performance of highly creative humans. Notably, even the top performing LLMs are still largely surpassed by highly creative individuals, underscoring a ceiling that current LLMs still fail to surpass. Our human-machine benchmarking framework addresses the polemic surrounding the imminent replacement of human creative labour by AI, disentangling the quality of the respective creative linguistic outputs using established objective measures. While prompting deeper exploration of the distinctive elements of human inventive thought compared to those of AI systems, we lay out a series of techniques to improve their outputs with respect to semantic diversity, such as prompt design and hyper-parameter tuning.
Raul P Pelaez, Guillem Simeon, Raimondas Galvelis, Antonio Mirarchi, Peter Eastman, Stefan Doerr, Philipp Tholke, Thomas E Markland, Gianni De Fabritiis
Journal of Chemical Theory and Computation | 2024
Achieving a balance between computational speed, prediction accuracy, and universal applicability in molecular simulations has been a persistent challenge. This paper presents substantial advancements in TorchMD-Net software, a pivotal step forward in the shift from conventional force fields to neural network-based potentials. The evolution of TorchMD-Net into a more comprehensive and versatile framework is highlighted, incorporating cutting-edge architectures such as TensorNet. This transformation is achieved through a modular design approach, encouraging customized applications within the scientific community. The most notable enhancement is a significant improvement in computational efficiency, achieving a very remarkable acceleration in the computation of energy and forces for TensorNet models, with performance gains ranging from 2× to 10× over previous, nonoptimized, iterations. Other …
Yassine El Ouahidi, Jonathan Lys, Philipp Thölke, Nicolas Farrugia, Bastien Pasdeloup, Vincent Gripon, Karim Jerbi, Giulia Lioi
arXiv preprint arXiv:2510.21585 | 2025
Foundation models have transformed AI by reducing reliance on task-specific data through large-scale pretraining. While successful in language and vision, their adoption in EEG has lagged due to the heterogeneity of public datasets, which are collected under varying protocols, devices, and electrode configurations. Existing EEG foundation models struggle to generalize across these variations, often restricting pretraining to a single setup, resulting in suboptimal performance, in particular under linear probing. We present REVE (Representation for EEG with Versatile Embeddings), a pretrained model explicitly designed to generalize across diverse EEG signals. REVE introduces a novel 4D positional encoding scheme that enables it to process signals of arbitrary length and electrode arrangement. Using a masked autoencoding objective, we pretrain REVE on over 60,000 hours of EEG data from 92 datasets spanning 25,000 subjects, representing the largest EEG pretraining effort to date. REVE achieves state-of-the-art results on 10 downstream EEG tasks, including motor imagery classification, seizure detection, sleep staging, cognitive load estimation, and emotion recognition. With little to no fine-tuning, it demonstrates strong generalization, and nuanced spatio-temporal modeling. We release code, pretrained weights, and tutorials to support standardized EEG research and accelerate progress in clinical neuroscience.
Philipp Thölke, Maxine Arcand-Lavigne, Tarek Lajnef, Sonia Frenette, Julie Carrier, Karim Jerbi
Communications Biology | 2025
Caffeine is the most widely consumed psychoactive stimulant worldwide. Yet important gaps persist in understanding its effects on the brain, especially during sleep. We analyzed sleep electroencephalography (EEG) in 40 subjects, contrasting 200 mg of caffeine against a placebo condition, utilizing inferential statistics and machine learning. We found that caffeine ingestion led to an increase in brain complexity, a widespread flattening of the power spectrum’s 1/f-like slope, and a reduction in long-range temporal correlations. Being most prominent during non-rapid eye movement (NREM) sleep, these results suggest that caffeine shifts the brain towards a critical regime and more diverse neural dynamics. Interestingly, this was more pronounced in younger adults (20–27 years) compared to middle-aged participants (41–58 years) during rapid eye movement (REM) sleep, while no significant age effects were …
Philipp Thölke, Antoine Bellemare-Pepin, Yann Harel, François Lespinasse, Karim Jerbi
Proceedings of the 15th International Conference on Computational Creativity | 2024
This paper introduces the” Bio-Mechanical Poet”, an adaptive brain-computer interface that integrates realtime electroencephalography (EEG) data with advanced generative artificial intelligence to create immersive audiovisual poetic experiences. We describe a custom prototyping environment for the exploration of various biosignals and their integration in a multimodal pipeline. By mapping brain states to symbolic representations, we explore trajectories of neural states in a multimodal symbolic latent space. This enables humaninterpretable access to it via the modalities of generative music, diffusion-based visuals and AI-crafted poetry. In doing so, we illustrate how the symbiosis of biosignals and generative systems can provide rich multimodal artworks guiding the user throughout the experience. Our discussion centers on the influence of biofeedback systems integrated with generative AI on evolving storytelling methods and altering perceptual states. We further discuss how translating biosignals into tangible expressions could open new avenues for understanding and interacting with our physiological and subconscious selves. Bio-Mechanical Poet exemplifies the potential of biofeedback and real-time feedback systems to foster advancements in the field of computational creativity, offering insights into the integration of human brain dynamics with artistic creation.
Annalisa Pascarella, Philipp Thölke, David Meunier, Jordan O’byrne, Tarek Lajnef, Antonino Raffone, Roberto Guidotti, Vittorio Pizzella, Laura Marzetti, Karim Jerbi
Neuroscience of Consciousness | 2025
While the beneficial impacts of meditation are increasingly acknowledged, its underlying neural mechanisms remain poorly understood. We examined the electrophysiological brain signals of expert Buddhist monks during two established meditation methods known as Samatha and Vipassana, which employ focused attention and open-monitoring technique. By combining source-space magnetoencephalography with advanced signal processing and machine learning tools, we provide an unprecedented assessment of the role of brain oscillations, complexity, and criticality in meditation. In addition to power spectral density, we computed long-range temporal correlations (LRTC), deviation from criticality coefficient (DCC), Lempel–Ziv complexity, 1/f slope, Higuchi fractal dimension, and spectral entropy. Our findings indicate increased levels of neural signal complexity during both meditation practices compared …
Carles Navarro, Mariona Torrens, Philipp Thölke, Stefan Doerr, Gianni De Fabritiis
arXiv preprint arXiv:2510.17826 | 2025
Building a working mental model of a protein typically requires weeks of reading, cross-referencing crystal and predicted structures, and inspecting ligand complexes, an effort that is slow, unevenly accessible, and often requires specialized computational skills. We introduce \emph{Speak to a Protein}, a new capability that turns protein analysis into an interactive, multimodal dialogue with an expert co-scientist. The AI system retrieves and synthesizes relevant literature, structures, and ligand data; grounds answers in a live 3D scene; and can highlight, annotate, manipulate and see the visualization. It also generates and runs code when needed, explaining results in both text and graphics. We demonstrate these capabilities on relevant proteins, posing questions about binding pockets, conformational changes, or structure-activity relationships to test ideas in real-time. \emph{Speak to a Protein} reduces the time from question to evidence, lowers the barrier to advanced structural analysis, and enables hypothesis generation by tightly coupling language, code, and 3D structures. \emph{Speak to a Protein} is freely accessible at https://open.playmolecule.org.
Antoine Bellemare-Pepin, Philipp Thölke, Yann Harel, Karim Jerbi
--- | 2024
In this paper, we present a novel system that integrates real-time neurofeedback into the creative process of generative AI, enabling seamless interactions between users and AI systems. By leveraging the user’s cognitive variability, the system allows for continuous and fluid co-creation, moving beyond the traditional promptbased interactions common in generative AI workflows. We achieve this using electroencephalography (EEG) to continuously monitor the user’s brain activity, which then acts as a control signal for a visual generative AI model. We focus specifically on Lempel-Ziv complexity, a measure of signal diversity that have previously been associated with mental states, task engagement and phenomenological richness. The proposed architecture includes an EEG feature extractor and a generative AI pipeline, working in tandem to dynamically alter the visual content of a pre-existing movie based on the user’s brain activity. This approach offers a new dimension of complexity and complicity in the interaction between humans and AI. Future work will explore the integration of more sophisticated bio-signals and multi-modal feedback, aiming to further enhance the depth and richness of the embodied creative experience. This work serves as a proof of principle for integrating biotechnology and generative AI in the emerging field of adaptive cinema. A playlist with video illustrations of the system in action can be found at youtube. com/playlist? list= PLMu36WzSQKiVeBnrUdwUAoUhqLqGX3_bw.
Antoine Bellemare-Pepin, Philipp Thölke, Karim Jerbi, Suzanne Kite
Proceedings of the Conference on Animation and Interactive Art, 89-98, 2025 | 2025
Oneiris is an interactive, AI-augmented brain-computer interface installation that explores personal and collective dreamscapes through generative artificial intelligence, real-time electroencephalography (EEG) neurofeedback, and Indigenous symbolic systems. Participants wear a wireless EEG headset and contribute dream narratives and hand-drawn sketches on a digital tablet. These inputs are embedded using Contrastive Language–Image Pre-training (CLIP) and matched to ten Lakota dream symbols, displayed as a floating constellation within a 360° projection space. A diffusion-based AI pipeline simultaneously augments participants’ sketches and texts into continuously evolving “dreamscapes”, whose texture and color palette are modulated in real time by neural markers of hypnagogia and brain complexity. A Medicine Wheel–inspired interface—an Indigenous symbol embodying the cyclical nature of life …