CFCedric Foucault

Cedric Foucault

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Cedric Foucault has a total of 13 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Adaptive learning in humans, brains and neural networks : the role of uncertainty and probabilities

    by CédRic Foucault

    Apprentissage adaptatif chez les humains, dans le cerveau et les réseaux de neurones : le rôle de l'incertitude et des probabilités: Lors de l’apprentissage, le cerveau est confronté à un problème difficile. Il apprend d’un environnement qui est à la fois stochastique et dynamique, c’est-à-dire qui se caractérise par des distributions de probabilité qui sont susceptibles de changer. La stochasticité incite le cerveau à intégrer un grand nombre d’observations passées pour arriver à une meilleure précision des connaissances acquises, tandis que la propension au changement l’incite à ne prendre en compte que les observations les plus récentes. Cette tension donne lieu à un compromis dans le taux d’apprentissage à utiliser, que le cerveau devrait idéalement adapter en permanence. Comment le cerveau parvient-il à un tel apprentissage adaptatif ? Selon la théorie normative, il devrait constamment représenter l’incertitude de ses connaissances par des distributions de probabilité, et les mettre à jour en pondérant les différentes probabilités, comme décrit dans les calculs de l’inférence bayésienne dynamique (aussi appelée filtrage bayésien). Guidés par cette théorie, mes travaux de thèse ont examiné les bases computationnelles et neurales de l’apprentissage adaptatif chez les humains à travers trois études. Dans la première étude (Foucault & Meyniel, soumise pour publication), j’ai examiné le comportement humain lors de l’apprentissage des magnitudes et des probabilités et testé son alignement avec la théorie normative. Grâce à un nouveau paradigme expérimental, j’ai mesuré les dynamiques du taux d’apprentissage des humains observation après observation, et les ai mises en relation avec les facteurs computationnels qui régissent les adaptations du taux d’apprentissage selon la théorie normative. J’ai montré que les humains adaptent leur taux d’apprentissage de façon dynamique et normative, guidés par l’incertitude, qui joue un rôle prépondérant dans l’apprentissage des probabilités. Globalement, j’ai constaté que le comportement humain suivait de près la théorie normative, offrant ainsi un fondement empirique solide pour ma thèse générale. Comment l’apprentissage adaptatif normatif observé chez les humains peut-il être réalisé dans le cerveau au niveau computationnel et mécanistique ? Dans la deuxième étude, théorique (Foucault & Meyniel, 2021, publiée dans eLife), j’ai exploré cette question en étudiant les mécanismes computationnels qui permettent à des réseaux de neurones de réaliser des capacités d’apprentissage adaptatif normatives. J’ai montré qu’avec un simple mécanisme de gating, qui pourrait être implémenté dans le cerveau par le système locus coeruleus-norepinephrine, un très petit réseau de neurones récurrent peut réaliser un apprentissage bayésien quasi-optimal. Ce mécanisme est essentiel pour la capacité du réseau à évaluer et utiliser l’incertitude pour adapter son taux d’apprentissage, comme je l’ai montré en décodant et en manipulant l’incertitude représentée dans les activités dynamiques récurrentes internes au réseau, et il est aussi essentiel pour d’autres capacités adaptatives. Une quantité fondamentale de notre théorie est la probabilité. Comme l’a montré la première étude, les humains sont clairement capables d’estimer la probabilité, mais comment est-elle représentée dans le cerveau ? Au niveau neural, la représentation de la probabilité est restée insaisissable. Ma troisième étude (Foucault*, Bounmy* et al., en préparation) tente de répondre à cette question en utilisant une nouvelle méthode alliant des modèles d’encodage à la théorie de l’approximation afin de modéliser le codage neural de la probabilité sans faire de supposition sur la forme du code. Cette méthode a révélé un code neural de la probabilité jusqu’alors inconnu dans l’activité cérébrale humaine mesurée par IRMf lors de l’apprentissage dans les cortex préfrontal dorsolatéral et intrapariétal. J’ai caractérisé ce code neural en reconstruisant ses tuning curves (fonctions d’encodage). J’ai ainsi constaté que, par rapport au code neural de l’incertitude

  • Open Access

    Pupil Dynamics Reflect Uncertainty-Driven Adjustments of Probability Learning

    by Aaron Greenhouse-Tucknott, Cedric Foucault, Anne Buot, Florent Meyniel

    Adaptive learning in dynamic environments depends on the brain’s ability to represent and respond to uncertainty. Changes in arousal-mediated pupil diameter provide insight into the neural basis of this process. Prior work has focused on magnitude-based inference tasks, and revealed that learning is primarily influenced by the detection of change points and the corresponding modulation of arousal. In comparison, the role of arousal systems in learning contexts where changes are harder to detect, such as in probability learning, remains poorly understood. Here, we combined pupillometry with a probability learning task in which human participants estimated a hidden generative probability of stimuli that changed abruptly and unpredictably. We adopted a Bayesian ideal observer model to estimate, trial-by-trial, the optimal apparent learning rate and the dynamics of two uncertainty-associated factors - namely, change point probability (the probability that the generative process has abruptly changed) and prior uncertainty (about the current beliefs on this generative process). In line with normative theory, participants heavily relied on prior uncertainty to adjust apparent learning rates. Pupil analyses revealed temporally distinct profiles that separate uncertainty factors: phasic dilations tracked change point probability while tonic pupil dilations tracked prior uncertainty. Mediation analyses further indicated that both phasic and tonic pupil signals at least partially carried the effects of change point probability and prior uncertainty on apparent learning rates, respectively. Together, our results demonstrate that probability learning in a dynamic environment is underpinned by computationally rational integration of latent uncertainty factors, implemented through arousal-associated responses.


  • Open Access

    Environmental dynamics shape human learning: change points versus random walks

    by Cedric Foucault, Lilian A Weber, Laurence Hunt

    Adaptive learning requires interpreting prediction errors in light of environmental dynamics. However, environments may not only differ in terms of when changes occur, but also how they occur: some nonstationary processes in nature exhibit slow drifts over time, while others show more abrupt changes. This raises the question as to whether humans can adapt their learning to reflect the generative structure of their environment. Here, we compared how humans learn in two canonical nonstationary environments: abrupt change points versus gradual random walks. Using a predictive inference task and a unified Bayesian framework, we show across two behavioral experiments that humans adapt their learning normatively between these two environments. Notably, identical prediction errors were interpreted differently across them. Large prediction errors triggered sharp increases in learning rate under change-point but not random-walk dynamics, where small and large errors were more equally weighted. This matched the predictions of a normative Bayesian model, which itself adopted the appropriate generative model for each environment. In addition, we found that humans could dissociate two latent variables that needed to be jointly inferred: the mean and variance (or stochasticity). This was demonstrated by explicit uncertainty reports that closely tracked the current variance, and only showed sustained changes following change points in variance but not mean. Together, these results show that humans adapt their learning strategy to both how and what aspects of the environment are changing. They establish a unified computational account of adaptive learning across different environmental dynamics.

  • Open Access

    Environmental dynamics shape human learning: change points versus random walks

    by Cedric Foucault, Lilian A. Weber, Laurence Hunt

    Adaptive learning requires interpreting prediction errors in light of environmental dynamics. However, environments may not only differ in terms of when changes occur, but also how they occur: some nonstationary processes in nature exhibit slow drifts over time, while others show more abrupt changes. This raises the question as to whether humans can adapt their learning to reflect the generative structure of their environment. Here, we compared how humans learn in two canonical nonstationary environments: abrupt change points versus gradual random walks. Using a predictive inference task and a unified Bayesian framework, we show across two behavioral experiments that humans adapt their learning normatively between these two environments. Notably, identical prediction errors were interpreted differently across them. Large prediction errors triggered sharp increases in learning rate under change-point but not random-walk dynamics, where small and large errors were more equally weighted. This matched the predictions of a normative Bayesian model, which itself adopted the appropriate generative model for each environment. In addition, we found that humans could dissociate two latent variables that needed to be jointly inferred: the mean and variance (or stochasticity). This was demonstrated by explicit uncertainty reports that closely tracked the current variance, and only showed sustained changes following change points in variance but not mean. Together, these results show that humans adapt their learning strategy to both how and what aspects of the environment are changing. They establish a unified computational account of adaptive learning across different environmental dynamics.


  • Open Access

    A non-monotonic code for event probability in the human brain

    by Cedric Foucault, Tiffany Bounmy, Sébastien Demortain, Bertrand Thirion, Evelyn Eger, Florent Meyniel

    Assessing probabilities and predicting future events are fundamental for perception and adaptive behavior, yet the neural representations of probability remain elusive. While previous studies have shown that neural activity in several brain regions correlates with probability-related factors such as surprise and uncertainty, similar correlations have not been found for probability. Here, using 7 Tesla functional magnetic resonance imaging, we uncover a representation of the probability of the next event in a sequence within the human dorsolateral prefrontal and intraparietal cortices. Crucially, univariate and multivariate analyses revealed that this representation employs a highly non-monotonic code. Tuning curves for probability exhibit selectivity to various probability ranges, while the code for confidence accompanying these estimates is predominantly monotonic. Given such diversity in tuning curves, future studies should move from assuming monotonic or simple canonical forms of tuning curves to considering richer representations, and clarify why different types of code exist.


    • 18 November 2025
    • Published by HAL in HAL
    Open Access

    A non-monotonic code for event probability in the human brain

    by Cedric Foucault, Tiffany Bounmy, Sébastien Demortain, Bertrand Thirion, Evelyn Eger, Florent Meyniel

    Assessing probabilities and predicting future events are fundamental for perception and adaptive behavior, yet the neural representations of probability remain elusive. While previous studies have shown that neural activity in several brain regions correlates with probability-related factors such as surprise and uncertainty, similar correlations have not been found for probability. Here, using 7 Tesla functional magnetic resonance imaging, we uncover a representation of the probability of the next event in a sequence within the human dorsolateral prefrontal and intraparietal cortices. Crucially, univariate and multivariate analyses revealed that this representation employs a highly non-monotonic code. Tuning curves for probability exhibit selectivity to various probability ranges, while the code for confidence accompanying these estimates is predominantly monotonic. Given such diversity in tuning curves, future studies should move from assuming monotonic or simple canonical forms of tuning curves to considering richer representations, and clarify why different types of code exist.

  • Open Access

    Two Determinants of Dynamic Adaptive Learning for Magnitudes and Probabilities

    by Cedric Foucault, Florent Meyniel

    Humans face a dynamic world that requires them to constantly update their knowledge. Each observation should influence their knowledge to a varying degree depending on whether it arises from a stochastic fluctuation or an environmental change. Thus, humans should dynamically adapt their learning rate based on each observation. Although crucial for characterizing the learning process, these dynamic adjustments have only been investigated empirically in magnitude learning. Another important type of learning is probability learning. The latter differs from the former in that individual observations are much less informative and a single one is insufficient to distinguish environmental changes from stochasticity. Do humans dynamically adapt their learning rate for probabilities? What determinants drive their dynamic adjustments in magnitude and probability learning? To answer these questions, we measured the subjects’ learning rate dynamics directly through real-time continuous reports during magnitude and probability learning. We found that subjects dynamically adapt their learning rate in both types of learning. After a change point, they increase their learning rate suddenly for magnitudes and prolongedly for probabilities. Their dynamics are driven differentially by two determinants: change-point probability, the main determinant for magnitudes, and prior uncertainty, the main determinant for probabilities. These results are fully in line with normative theory, both qualitatively and quantitatively. Overall, our findings demonstrate a remarkable human ability for dynamic adaptive learning under uncertainty, and guide studies of the neural mechanisms of learning, highlighting different determinants for magnitudes and probabilities.


  • Open Access

    A nonlinear code for event probability in the human brain

    by Cedric Foucault, Tiffany Bounmy, Sébastien Demortain, Bertrand Thiririon, Evelyn Eger, Florent Meyniel

    Assessing probabilities and predicting future events are fundamental for perception and adaptive behavior, yet the neural representations of probability remain elusive. While previous studies have shown that neural activity in several brain regions correlates with probability-related factors such as surprise and uncertainty, similar correlations have not been found for probability. Here, using 7 Tesla functional magnetic resonance imaging, we uncover a representation of the probability of the next event in a sequence within the human dorsolateral prefrontal and intraparietal cortices. Crucially, univariate and multivariate analyses revealed that this representation employs a highly nonlinear code. Tuning curves for probability exhibit selectivity to various probability ranges, while the code for confidence accompanying these estimates is predominantly linear. The diversity of tuning curves we found recommends that future studies move from assuming linear correlates or simple canonical forms of tuning curves to considering richer representations whose benefits remain to be discovered.


  • Open Access

    Two determinants of dynamic adaptive learning for magnitudes and probabilities

    by Cedric Foucault, Florent Meyniel

    Humans face a dynamic world that requires them to constantly update their knowledge. Each observation should influence their knowledge to a varying degree depending on whether it arises from a stochastic fluctuation or an environmental change. Thus, humans should dynamically adapt their learning rate based on each observation. Although crucial for characterizing the learning process, these dynamic adjustments have only been investigated empirically in magnitude learning. Another important type of learning is probability learning. The latter differs from the former in that individual observations are much less informative and a single one is insufficient to distinguish environmental changes from stochasticity. Do humans dynamically adapt their learning rate for probabilities? What determinants drive their dynamic adjustments in magnitude and probability learning? To answer these questions, we measured the subjects’ learning rate dynamics directly through real-time continuous reports during magnitude and probability learning. We found that subjects dynamically adapt their learning rate in both types of learning. After a change point, they increase their learning rate suddenly for magnitudes and prolongedly for probabilities. Their dynamics are driven differentially by two determinants: change-point probability, the main determinant for magnitudes, and prior uncertainty, the main determinant for probabilities. These results are fully in line with normative theory, both qualitatively and quantitatively. Overall, our findings demonstrate a remarkable human ability for dynamic adaptive learning under uncertainty, and guide studies of the neural mechanisms of learning, highlighting different determinants for magnitudes and probabilities. Significance statement: In a dynamic world, we must constantly update our knowledge based on the observations we make. However, how much should we update our knowledge after each observation? Here, we have demonstrated two principles in humans that govern their updating and by which they are capable of dynamic adaptive learning. The first principle is that when they observe a highly surprising event indicating a likely change in the environment, humans reset their knowledge and perform one-shot learning. The second principle is that when their knowledge is more uncertain, humans update it more quickly. We further found that these two principles are differentially called upon in two key learning contexts that could be associated with different brain mechanisms: magnitude learning (which primarily requires adaptation to surprise, under the first principle) and probability learning (which primarily requires adaptation to uncertainty, under the second principle). Our findings advance understanding of the mechanisms of human learning, with implications for the brain and the development of adaptive machines.


  • Open Access

    Gated recurrence enables simple and accurate sequence prediction in stochastic, changing, and structured environments

    by Cédric Foucault, Florent Meyniel

    From decision making to perception to language, predicting what is coming next is crucial. It is also challenging in stochastic, changing, and structured environments; yet the brain makes accurate predictions in many situations. What computational architecture could enable this feat? Bayesian inference makes optimal predictions but is prohibitively difficult to compute. Here, we show that a specific recurrent neural network architecture enables simple and accurate solutions in several environments. This architecture relies on three mechanisms: gating, lateral connections, and recurrent weight training. Like the optimal solution and the human brain, such networks develop internal representations of their changing environment (including estimates of the environment’s latent variables and the precision of these estimates), leverage multiple levels of latent structure, and adapt their effective learning rate to changes without changing their connection weights. Being ubiquitous in the brain, gated recurrence could therefore serve as a generic building block to predict in real-life environments.