Real neurons, and their networks, are far too complex to be described exactly by simple deterministic equations. Any description of their dynamics must therefore incorporate noise to some degree. It is my thesis that the nervous system is organized in such a way that its performance is optimal, subject to this constraint. I further contend that neuronal dynamics may even be enhanced by noise, when compared with their deterministic counter-parts. To support my thesis I will present and analyze three case studies. I will show how noise might (i) extend the dynamic range of mammalian cold-receptors and other cells that exhibit a temperature-dependent discharge; (ii) feature in the perception of ambiguous figures such as the Necker cube; (iii) alter the discharge pattern of single cells.

When viewing ambiguous figures, individuals can exert selective attentional control over their perceptual reversibility behaviour (e.g., Strüber & Stadler, 1999). In the current study, we replicated this finding but we also found that ambiguous figures containing faces are processed quite differently from those containing objects. Furthermore, inverting an ambiguous figure containing faces (i.e., Rubin’s vase-face) resulted in an “inversion effect”. These findings highlight the importance of considering how we attend to faces in addition to how we perceive and process faces. Describing the perceptual reversal patterns of individuals in the general population allowed us to draw comparisons to behaviours exhibited by individuals with Asperger Syndrome (AS). The group data suggested that these individuals were less affected by figure type or stimulus inversion. Examination of individual scores, moreover, revealed that the majority of participants with AS showed an atypical reversal pattern, particularly with ambiguous figures containing faces, and an atypical inversion effect. Together, our results show that ambiguous figures can be a very valuable tool for examining face processing mechanisms in the general population and other distinct groups of individuals, particularly those diagnosed with AS. / October 2008

When viewing ambiguous figures, individuals can exert selective attentional control over their perceptual reversibility behaviour (e.g., Strüber & Stadler, 1999). In the current study, we replicated this finding but we also found that ambiguous figures containing faces are processed quite differently from those containing objects. Furthermore, inverting an ambiguous figure containing faces (i.e., Rubin’s vase-face) resulted in an “inversion effect”. These findings highlight the importance of considering how we attend to faces in addition to how we perceive and process faces. Describing the perceptual reversal patterns of individuals in the general population allowed us to draw comparisons to behaviours exhibited by individuals with Asperger Syndrome (AS). The group data suggested that these individuals were less affected by figure type or stimulus inversion. Examination of individual scores, moreover, revealed that the majority of participants with AS showed an atypical reversal pattern, particularly with ambiguous figures containing faces, and an atypical inversion effect. Together, our results show that ambiguous figures can be a very valuable tool for examining face processing mechanisms in the general population and other distinct groups of individuals, particularly those diagnosed with AS.

When viewing ambiguous figures, individuals can exert selective attentional control over their perceptual reversibility behaviour (e.g., Strüber & Stadler, 1999). In the current study, we replicated this finding but we also found that ambiguous figures containing faces are processed quite differently from those containing objects. Furthermore, inverting an ambiguous figure containing faces (i.e., Rubin’s vase-face) resulted in an “inversion effect”. These findings highlight the importance of considering how we attend to faces in addition to how we perceive and process faces. Describing the perceptual reversal patterns of individuals in the general population allowed us to draw comparisons to behaviours exhibited by individuals with Asperger Syndrome (AS). The group data suggested that these individuals were less affected by figure type or stimulus inversion. Examination of individual scores, moreover, revealed that the majority of participants with AS showed an atypical reversal pattern, particularly with ambiguous figures containing faces, and an atypical inversion effect. Together, our results show that ambiguous figures can be a very valuable tool for examining face processing mechanisms in the general population and other distinct groups of individuals, particularly those diagnosed with AS.

Nous évoluons dans un monde incertain. De ce fait, notre survie dépend de notre capacité à prendre rapidement des décisions, et ce de manière fiable et adaptative. Il est possible de mieux comprendre cette capacité en considérant la perception comme un processus d’inférence probabiliste au cours duquel les informations sensorielles sont combinées à nos attentes pour produire une interprétation plausible de notre environnement. Les théories récentes de psychiatrie computationnelle suggèrent par ailleurs que la grande variabilité des troubles psychiatriques, au rang desquelles figure la schizophrénie, pourrait résulter d’une altération de ces mêmes processifs prédictifs. L’Inférence Circulaire est l’une de ces théories. Ce cadre de pensée stipule qu’une propagation incontrôlée d’information dans la hiérarchie corticale pourrait générer des percepts ou des croyances aberrantes. Afin d’explorer le rôle joué par l’Inférence Circulaire en condition normale ou pathologique, ce travail de thèse s’est appuyé sur des tâches de prise de décision en conditions perceptives ambigües. Dans une première partie, nous nous sommes intéressés au rôle joué par la circularité dans la perception bistable. Le phénomène de bistabilité survient lorsque deux interprétations se succèdent à intervalle régulier pour un même percept. Nous présentons les résultats d’une tâche conduite en population saine où nous avons manipulé les informations sensorielles et à priori utilisées par les participants lors de la visualisation d’un cube de Necker (article 1). Nous avons pu montrer un effet propre à chaque manipulation, mais également une interaction entre ces deux sources d’information, incompatible avec une intégration Bayésienne optimale. Résultat confirmé par la comparaison de divers modèles computationnels ajustés aux données, qui a pu mettre en évidence la supériorité de l’Inférence Circulaire sur les modèles Bayésiens classiques. Nous avons ensuite voulu tester un modèle fonctionnel de la bistabilité (article 2). Nous avons donc dérivé la dynamique du modèle et montré que la présence de boucles descendantes dans la hiérarchie corticale, transformait ce qui était jusque là un intégrateur imparfait du bruit sensoriel en modèle à attracteur bistable. Ce modèle ne reproduit pas seulement le phénomène de bistabilité, mais également l’ensemble de ces caractéristiques phénoménologiques. Dans un 3ème article, nous avons testé une prédiction, notamment en cas de présentation discontinue d’un stimulus bistable. Deux expériences complémentaires utilisant un paradigme de présentation intermittente du cube de Necker ont donc été conduites en population générale. Nos résultats étaient compatible avec les prédictions faites par le modèle de l’Inférence Circulaire Dynamique, suggérant que la circularité puisse être un mécanisme générique à l’origine de notre façon de voir le monde. Dans la seconde partie de ce travail, nous avons étudié l’Inférence Circulaire en condition pathologique, notamment lors d’expériences psychotiques (schizophrénie, psychédéliques). Nous avons utilisé la perception bistable pour explorer les mécanismes computationnels à l’œuvre dans la schizophrénie (article 4,5). Nous avons comparé les performances de patients présentant des symptômes psychotiques à des témoins sains appariés lors d’une tâche de perception bistable. Nous avons pu montrer chez les patients une amplification des informations sensorielles combinée à une surestimation de la volatilité environnementale. Enfin nous terminons ce travail en proposant une approche transversale de l’effet des psychédéliques (article 6), sur la base des résultats précédents et de la spécificité clinique de ces expériences sensorielles cross-modales, afin de relier l’échelle macroscopique (i.e., comportement et phénoménologie), mésoscopique (i.e., les boucles inférentielles) et microscopique (i.e., les différents neurotransmetteurs impliqués aboutissant à un microcircuit canonique). / We live in an uncertain world, yet our survival depends on how quickly and accurately we can make decisions and act upon them. To address this problem, modern neuroscience reconceptualised perception as an inference process, in which the brain combines sensory inputs and prior expectations to reconstruct a plausible image of the world. In addition to that, influential theories in the emerging field of computational psychiatry suggest that various psychiatric disorders, including schizophrenia, could be the outcome of impaired predictive processing. Among those theories, the circular inference framework suggests that an unconstrained propagation of information in the cortex, underlain by an excitatory to inhibitory imbalance, can generate false percepts and beliefs, similar to those exhibited by schizophrenia patients. In the present thesis, we probed the role of circular inference from normal to pathological brain functioning, gaining insights from perceptual decision making in the presence of high ambiguity. In the first part of the thesis, we focused on the role of circularity in bistable perception in the general population. Bistability occurs when two mutually exclusive interpretations compete and switch as dominant percepts every few seconds. In a 1st article, we manipulated sensory evidence and priors in a Necker cube task, asking how the brain combines low-level and high-level information to form perceptual interpretations. We found a significant effect of each manipulation but also an interaction between the two, a finding incompatible with Bayes optimal integration. Bayesian model comparison further supported this observation, showing that a circular inference model outperformed purely Bayesian models. Having established a link between circular inference and bistable perception, we then put forward a functional theory of bistability, based on circularity (2nd article). In particular, we derived the dynamics of a dynamical circular inference model, showing that descending loops (i.e. a form of circularity resulting in aberrant amplification of the priors) transform what is normally a leaky integration of noisy evidence into a bistable attractor with two highly trusted stable states. Importantly, this model can explain both the existence and the phenomenological properties of bistable perception, making a number of testable predictions. Finally, in a 3rd article, we tested one of the model’s predictions, namely the perceptual behaviour when the stimulus is presented discontinuously. We ran two Necker cube experiments using a novel intermittent-presentation methodology, and we calculated the stabilisation curves (i.e. persistence as a function of blank durations). We found that participants’ behaviour was compatible with the model’s prediction for a system with descending loops, suggesting that circularity constitutes a general mechanism that shapes the way healthy individuals perceive the world. In the second part, we studied circular inference in pathological conditions related to psychosis. We notably focused on two varieties of the psychotic experience, namely schizophrenia-related psychosis and drug-induced psychosis. After discussing the links between behaviour, aberrant message-passing and the corresponding neural networks (4th article), we used bistable perception to probe the computational mechanisms underlying schizophrenia in a 5th article. We compared patients with prominent positive symptoms with matched healthy controls in two bistable perception tasks. Our results suggest an enhanced amplification of sensory inputs in patients, combined with an overestimation of the environmental volatility. In the last article (6th), we delineated a multiscale account of psychedelics, ultimately linking the macroscale (i.e. phenomenological considerations such as the crossmodal character of the psychedelics experience), the mesoscale (i.e. loops) and the microscale (i.e. neuromodulators and canonical microcircuits).