[Risc] bibliothèque : Mémoires du Cogmaster 2021 Boîte de réception Pourinfos Se désabonner mer. 22 sept. 14:24 (il y a 21 heures) À Bonjour, Les mémoires du Cogmaster soutenus en 2021 sont disponibles au Risc. En voici la liste. Bonne lecture, L’équipe du Risc ________________________ • ANTOINE Victor Quantifying near-homophony induced by French liaison Sous la direction de Sharon Peperkamp et Rory Turnbull Résumé :Liaison in French consists of the pronunciation of an otherwise silent underlying word-final consonant when the following word is vowel-inital e.g. dernier chat 'last cat' [d??.nje.?a] but dernier achat 'last purchase' [d??.nje.?a.?a]. Liaison consonants can lead to near-homophonous sequences e.g. dernier rachat 'last repurchase' is pronouced as dernier achat. Such sequences raise difficulties as speakers activate both forms - but must be able to determine which one is intended - and thus are processed more slowly than non-homophonous ones (Spinelli et al., 2003). We investigated whether French is structured in such a way as to reduce liaison-induced near- homophonous environments. Using the corpus Lexique, we extracted (i) all words with an underlying liaison consonant e.g. dernier and (ii) all pairs of words differing only by the presence vs. absence of an initial consonant e.g. achat / rachat. We combined them to create confusing doublets e.g. {dernier achat / d! ernier rachat} – keeping only grammatical ones. As we were interested in whether liaison consonants were the ones that produced the least number of doublets, we devised substitutions of liaison consonants. Words were identical except for their liaison consonant - e.g. dernier would have [f] and no longer [?] as liaison consonant – and thus giving new confusing doublets – e.g. {dernier acteur 'last actor' / dernier facteur 'last factor'}. We found a significantly lower number of confusing doublets for four of the six liaison consonants. Post-hoc frequency analyses were run to test whether these doublets were troublesome in everyday speech. We selected so-called troublesome doublets i.e. (i) those which the speaker experienced frequently and (ii) which were difficult to segment. We found a significantly lower number of troublesome doublets for three of the six liaison consonants. We claim that these results suggest that French is indeed structured such as to minimize li! aison-induced near-homophony. Mots-clés : Liaison; Homophones; Lexicon; Lexical segmentation; French 24 p., 2021 Laboratoire de stage : LSCP - Paris Spécialité : langage Cote pour l’emprunt : Master 1902 _____________________________ • AUDOUY Flore Confidence in having a conscious perception could track conscious perception in Hakwan Lau's Perceptual Reality Monitoring Sous la direction de Nathan Faivre Résumé : Morales & Lau, forthcoming, argue that the level of confidence in a detection task reflects the degree of perceptual awareness. Yet, their claim is undermined by cases where confidence is informed by prior beliefs. Indeed, in this case, confidence would not reflect the degree of conscious perception. However, Sandberg et al., 2010, highlight two different uses of confidence judgment in a detection task that Morales and Lau do not seem to disentangle. On one side, the participant can rate his confidence in having a phenomenal experience of the signal; on the other side, he can rate is confidence in having provided a correct perceptual response. Based on Lau's model of 'Perceptual Reality Monitoring' ('PRM') (Lau, 2019), this thesis proposes to investigate which of these two forms of confidence could track consciousness. Indeed, PRM asserts that the formation of conscious perception depends on a metacognitive system that monitors the quality of a first-order state. ! If the monitoring mechanism 'decides' that a perceptual evidence has been externally triggered, then a conscious perception with assertoric force is formed. Assertoric force is a phenomenal component of conscious perception that inclines the perceiver to believe his perceptual content. If the place of assertoric force in PRM seems to show that confidence in the correctness of the response tracks consciousness, the case of resisted hallucinations will suggest that it is not always true. Therefore, it will be argued that confidence in having a phenomenal experience of the signal might track consciousness. In particular, it will be argued that Perceptual Awareness Scales closely resemble to this latter form of confidence in PRM. Therefore, if the interpretation of Lau's model is correct, then it should be confidence in having a phenomenal experience of the signal that should track consciousness, and even if confidence is informed by prior beliefs. Mots-clés : Metacognition - Confidence judgment - Conscious perception – Reality Monitoring – Assertoric force 56 p., 2021 Laboratoire de stage : Laboratoire de psychologie et neurocognition, Grenoble Spécialité : Psychologie Cote pour l’emprunt : Master 1903 ______________________ • BAQUE Eugénie Métacognition de la schizophrénie : une étude comparant jugements de confiance en mémoire et perception Sous la direction de Paul Roux et Nathan Faivre Résumé : La métacognition se définit comme la fonction cognitive permettant d’évaluer et d’agir sur ses propres processus cognitifs, et ainsi de se représenter réflexivement soi-même. Des symptômes cliniques présents dans la schizophrénie tels que le manque d’insight ou les idées délirantes pourraient s’expliquer par un défaut d’ajustement de la confiance dans des processus perceptuels, mnésiques, ou par un déficit global en métacognition. Des résultats contradictoires existent dans la littérature, suggérant parfois l’existence d’un excès de confiance dans des erreurs commises lors de tâches cognitives, et parfois une absence de différence avec les sujets contrôles. L’intérêt de développer un modèle neurocognitif de la schizophrénie incluant la métacognition comme processus central au développement de la pathologie s’allie avec l’intérêt de développer des mesures de performances métacognitives plus précises, et permetta! nt de mettre en évidence les déficits s’ils existent, en s’affranchissant de la confusion avec les mesures de performances de premier ordre. Ce travail de recherche présente l’élaboration d’une tâche évaluant conjointement mémoire autobiographique, mémoire épisodique et perception, associés à un jugement de confiance, avec l’hypothèse de trouver un déficit métacognitif chez les patients et d’évaluer l’aspect global ou spécifique à un domaine de ce déficit. Nous avons pu faire passer cette tâche à quatre sujets contrôles ainsi qu’à quatre sujets présentant une schizophrénie. Les mesures de performance de type 1 et de type 2 ont été calculées en appliquant la théorie de détection du signal, l’efficience métacognitive mesurée par le M-ratio permettant d’avoir une mesure de performance métacognitive sans contamination venant du biais de confiance ou de la performance de premier ordre. Les recrutements n’ont pas été suffisa! mment nombreux pour permettre de comparer les patients aux témoins, et nous n’avons pas mis en évidence de déficit en métacognition. Mots-clés : 69 p., 2021 Laboratoire de stage : Centre de recherche en épidémiologie et santé des populations - Cesp, Versailles Spécialité : Neurosciences Cote pour l’emprunt : Master 1953 ______________________ • BAUDIER Laure La Flexibilité Cognitive : une compétence générale, dépendante du contexte ou d’une composante flexible spécifique? Sous la direction d’Evelyne Clément Résumé : La flexibilité cognitive est habituellement décrite comme une capacité générale, indépendante du contexte, se développant dès le plus jeune âge suite à la maturation du cerveau (Diamond, 2013 ; Zelazo & Frye, 1998). Dans cette perspective, diverses tâches et méthodes reposant sur le changement de tâche ont été conçues dans l'intention d'entraîner et d'évaluer cette capacité, mais comme le note Clément (2021), ces paradigmes ne font qu'évaluer la composante de changement de tâche des fonctions exécutives, et non toute l'étendue de ce qu'est réellement la flexibilité cognitive. Implicitement, ces tâches partent de l'hypothèse que la flexibilité est une aptitude générale au domaine qui repose sur le déplacement de l'attention entre deux tâches. Or, cette hypothèse fait l'objet d'un débat animé dans la littérature concernant la nature de la flexibilité cognitive. Par exemple, les résultats d'expériences utilisant deux formes de ! tâches de flexibilité structurellement similaires (basées sur un changement de règles ou d’inférences) suggèrent que la flexibilité cognitive n'est pas un trait globalement cohérent (Deák & Whiseheart, 2015). De plus, des études neuropsychologiques ont démontré l'existence de deux formes de flexibilité cognitive : la flexibilité réactive et la flexibilité spontanée. La flexibilité réactive désigne la capacité à passer d'un stimulus ou d'une tâche en réponse à des contraintes environnementales, tandis que la flexibilité spontanée décrit la capacité à produire une diversité d'idées ou adopter différentes perspectives. Nous avons étudié la nature de la flexibilité cognitive chez des enfants de ** et analysé leurs performances individuelles à deux tâches de flexibilité spontanée, deux tâches de flexibilité adaptative et trois épreuves de fonctionnement exécutif. Les passations ayant été perturbées depuis ce début d’année 2021, ce mémoire rapporte des résultats préliminaires sur une population de 14 sujets qui ont complété l’ensemble des épreuves. Ces premières analyses ne sont pas vouées à être généralisées, mais les analyses de corrélations ainsi que la construction des modèles de prédicteurs exécutifs suggèrent que la flexibilité serait très sensible au contexte dans lequel elle s’exprime ainsi qu’ aux connaissances impliquées par chaque tâche. Mots-clés : Psychologie ; Felxibilité cognitive 34 p., 2021 Laboratoire de stage : Laboratoire Paragraphe - Cergy Paris Université Spécialité : Psychologie Cote pour l’emprunt : Master 1904 ______________________ • BEAUD Alice Effects of attention on cognitive performances in wild vervet monkey Sous la direction de Erica Van de Waal et Tecla Mohr Résumé : Face to daily challenges of surviving in the wild, individuals develop behavioural flexibility. Cognitive abilities play a major role in driving flexible behaviours as it is defined as processes by which animals collect, retain and use information from their environment. Testing cognitive capacities in the wild, implies to take in consideration many internal factors, usually controlled in captivity, that can influence cognitive performance. We investigated the influence of attention (measured as the head orientation) and of the time spent interacting with the experimental device on cognitive performance, by conducting the first associative learning task displayed on a touchscreen in the wild in free-ranging vervet monkeys (Chlorocebus pygerythrus). We showed that when the monkeys paid attention to screen, they made significantly more correct answers. High level of attention during the task was linked to many touches made with the muzzle. We found that a longer tim! e spent interacting with the screen was correlated with higher performance whereas the time spent looking at the screen was decreasing over sessions. These last results were depending on the performance level of the individuals. These results highlight the importance of considering confounding factors influence before concluding to difference in cognitive abilities between individuals and species. The increase focus on intra and inter-individual variations in performance will lead to a better understanding in the evolutionary driving forces on cognitive abilities. Mots-clés : Ethologie, attenetion, apprentissage 37 p., 2021 Laboratoire de stage : Faculté de biologie et de médecin, département d'écologie et évolution, Université de Lausanne Spécialité : Ethologie Cote pour l’emprunt : Master 1952 ______________________ • BEAUTÉ Romy Multicentric exploration of cognitive motor dissociations in unresponsive patients using high-density EEG Sous la direction de Benjamin Rohaut Résumé : Background Clinically unresponsive patients may show signs of residual consciousness when probed using brain imagery (e.g. fMRI, ERP, EEG, etc). In approximately 15% of unrespon- sive patients (i.e., in a comatose, vegetative or minimally conscious state), robust evi- dence of brain activations in response to a motor verbal commands can be found despite any corresponding reliable behavior. This state of dissociation between a lack of appar- ent behavioral response but a preservation of brain responses has recently been coined cognitive-motor dissociation (Cognitive Motor Dissociations (CMD)) ( (Schiff, 2015)). CMD can be detected using simple bed-side EEG active paradigms. Combined with machine learning methods, such approach provides a robust statistical tool to screen clinically unresponsive patients for covert awareness. However, understanding about their residual cognition and conscious content are still lacking. The goal of this master thesis was two-folds. A first part consisted ! in the replication, adaptation and standardization of a previously published paradigm, to detect CMD across 4 international centers. A second part was directed toward building a better understanding of the pathophysio- logical mechanisms and of the conscious experience of CMD state. Methods We recruited 21 healthy volunteers and included a series of 55 patients admitted in the intensive care unit Intensive Care Unit (ICU). Patients presented various etiologies of acute brain injuries (e.g., anoxia, traumatic, etc..) and were all clinically unrespon- sive to verbal spoken commands. EEG were recorded while delivering repetitive spoken verbal commands (i.e., "keep/stop opening and closing your right/left hand"). A super- vised classifier (SVM) was then fitted on individual EEG data and trained to distinguish "move" vs "rest" EEG periods. Diagnosis of CMD was defined by an AUC significantly greater than the chance level (AUC> 0.5 with a p-value < 0.05 on a permutation test) SUMMARY iwhen discriminating "move" vs "rest" EEG periods. Temporal and spatial patterns of classifier outputs as well as a source localisations of the power spectrum modulations of the EEG across conditions were computed for descriptive purpose and to better as- sess the neurophysiological underpinning of CMD. Finally, a standardized questionnaire was created to obtain robust verbal reports from CMD patients that would regained full functional communication. Results Detection of brain activation in response to motor commands were observed in 5 out of 55 patients (9%) and in 20 out of 21 (99%) healthy volunteers. A subset of healthy volunteers was assessed during an imaginary motor task for which a lower sensitivity was observed (5/11 , 45%). The classifiers’ performances were stable for a number of electrodes> 8. Spatial patterns and source localisations did not show motor contralateral topographies but possibly form the supplementary motor areas. Subjective experience of CMD could be assessed in one patient and showed no memory of the experiment although the patients could report some memories close to the recording period. Conclusion We set up an standardized pipeline that detected presence of covert responses in 9% of patients with acute brain injury, recorded across 4 internal centers. We obtained a very rich verbal report from a patient that recovered from a CMD state. The richness of this report suggest that brain activation in response to motor command are more than purely reflexive. 79 p., 2021 Laboratoire de stage : Physiological investigations of clinically normal and impaired cognition Lab (PICNIC Lab) - ICM Spécialité : Neurosciences Cote pour l’emprunt : Master 1951 ______________________ • BERGEROT Clémence Choice-confirmation bias in favorable vs. adverse environments Sous la direction de Valérian Chambon Résumé : When learning from their actions in order to make subsequent decisions, people tend to give more weight to positive outcomes than to negative outcomes. Recent evidence sug- gests that such optimism is a feature of self-determined actions, as opposed to imposed choices that would induce unbiased learning. Whether optimistic learning occurs when imposed outcomes are aligned with an agent’s preferences, however, is unknown. This the- sis investigates this question through the elaboration of an experimental task involving a "congruency" parameter, which quantifies the proportion of preference-aligned imposed outcomes in a trial sequence. Throughout this work, the impact of congruency on learn- ing is assessed by simulations of virtual agents on an experimental design in which other parameters are manipulated: type of choice (free vs. imposed), and outcome contingen- cies (highly rewarded vs. poorly rewarded). Results show that congruency’s impact on learning is mo! dulated by outcome contingencies and agents’ biases. In particular, optimal performance across outcome contingencies was obtained by agents whose biases were opti- mistic in free choices, optimistic in preference-aligned imposed choices, and pessimistic in preference-nonaligned imposed choices. These results lead to the design of an experimen- tal task which enables testing for the validity of this three-bias model. Moreover, they constitute a new step towards the understanding of optimistic biases in reinforcement learning and, in particular, of their adaptive character in various environments. Mots-clés : Reinforcement learning; Asymmetric value update; Confirmation bias; Learn- ing rate 56 p., 2021 Laboratoire de stage : Institut Jean Nicod, CNRS-EHESS-ENS, Paris Spécialité : Cote pour l’emprunt : Master 1950 ______________________ • BRUNO Nicolas Exploratory analysis of EEG markers of mind-wandering Sous la direction d’Antonin Valero-Cabre et Jacob Sitt Résumé : Mind-wandering is a universal mental phenomenon typically characterized as an atten- tion directed away from an external task towards internal cognition. To date, several studies have attempted to study the electroencephalographic (EEG) signature of mind-wandering (MW) without consistent findings. Moreover, recent studies have shown that the MW phe- nomenon is not homogeneous but rather fluctuates along several dimensions. One such di- mension, intentionality, has been found to be key in separating deliberate from spontaneous forms of MW. The present study explores which EEG markers were best for differentiating be- tween on-task and MW states. Also, we extended the same analysis to deliberate and sponta- neous MW. Our results suggest that the best marker to discriminate on-task states of attention from others was the power of oscillations in the theta band (4-7 Hz). Conversely, the most consistent measures to discriminate between deliberate and spontaneous MW, we! re alpha band (8-12 Hz) power measures. Importantly, by comparing across highest ranked markers we found that none were discriminative when employed for the classification of other states of attention. These results further confirm the heterogeneous nature of MW and the distinctive- ness of on-task states and of deliberate and spontaneous MW states. Therefore, these states are likely to not only represent different experiential features of cognition but also to reflect different neural underpinnings. Mots-clés : Mind-wandering, task unrelated thought, EEG, deliberate, spontaneous 84 p., 2021 Laboratoire de stage : Frontlab et PICNIC Lab, Institut du cerveau (ICM) Spécialité : Neurosciences Cote pour l’emprunt : Master 1949 ______________________ • CARBUCCIA Laudine Les barrières d’accès aux modes d’accueil formels chez les populations défavorisées : une approche comportementale Sous la direction de Coralie Chevallier et Carlo Baron Résumé : Alors que les modes d’accueil formels (c’est-à-dire principalement les crèches et les assistantes maternelles) sont des leviers d’action très efficaces pour réduire les inégalités de développement présentes dès la petite enfance, ces structures sont elles-mêmes marquées par de fortes inégalités d’accès. Les raisons de la sous-représentation des familles défavorisées ont déjà été étudiées, mais pas de manière systématique. Le travail de revue présenté dans ce mémoire a permis 1) de construire un modèle intégratif des barrières d’accès aux modes d’accueil formels pour ces populations 2) d’évaluer ce modèle à travers une revue systématique PRISMA de la littérature. Les barrières socio-structurelles sont les seules visées par les politiques publiques actuelles. Pourtant, notre revue met en évidence que des barrières de nature psychologique, qui n’ont jamais été théorisées jusqu’à présent, sont au moins a! ussi importantes que ces barrières socio-structurelles. De nouvelles politiques publiques devraient être créées pour agir sur les barrières psychologiques, faute de quoi l’efficacité des réformes structurelles pourrait s’avérer fortement limitée. Mots-clés : développement cognitif ; petite enfance ; inégalités socio-économiques ; modes d’accueil formels ; barrières d’accès ; revue systématique PRISMA ; p-curving. 79 p., 2021 Laboratoire de stage : Institut Jean Nicod - Paris Spécialité : Cognition sociale Cote pour l’emprunt : Master 1905 ______________________ • CLAVE Amélie Du laboratoire à la réalité : la prospect theory nous permet-elle de prédire les comportements des investisseurs ? Sous la direction de Valérian Chambon et Tiphaine Saltini Résumé : En 1979, Daniel Kahneman et Amos Tversky ont développé une nouvelle théorie en économie comportementale : la théorie des perspectives. Quarante ans plus tard, elle reste au cœur de nombreuses réflexions et débats pour chercher à comprendre et anticiper les comportements des agents économiques lors de la prise de décision dans un contexte de risque. De nombreux chercheurs continuent en effet à s’interroger sur la pertinence de cette théorie, et plus particulièrement sur la possibilité de l’appliquer à la réalité et de généraliser ses conclusions hors du laboratoire. Cette interrogation semble légitime puisque cette théorie a reçu de nombreuses validations théoriques mais qu’elle se heurte à quelques limites lorsqu’il s’agit d’extrapoler ses conclusions à la vie quotidienne et à la réalité hors du laboratoire. La méthodologie spécifique utilisée au sein des laboratoires pour les expériences, la question de la stabilité d! es préférences individuelles dans le temps ainsi que la découverte de plusieurs autres biais cognitifs qui semblent pouvoir apporter des subtilités dans une théorie de la prise de décision dans un contexte de risque sont autant d’éléments qui invitent à nuancer les conclusions et les ambitions de la théorie des perspectives. Face à ce constat et à ces freins, nous avons tenté d’identifier des pistes pour améliorer la fiabilité des résultats obtenus grâce à la théorie des perspectives et augmenter sa capacité prédictive : adapter les méthodes de capture des préférences à une population plus large et moins éduquée, se trouver dans une situation de risque, où les décisions sont prises à l’échelle individuelle, considérer les résultats comme des indicateurs statiques des préférences de risque et réfléchir à des méthodes alternatives pour agrandir les échantillons. Mots-clés : théorie des perspectives, économie comportementale, biais cognitifs 30 p., 2021 Laboratoire de stage : Spécialité : Economie expériemntale Cote pour l’emprunt : Master 1906 ______________________ • D'AMELIO Tomás Ariel Predicting modeling of arousal: an analysis of a public EEG and EDA database Sous la direction de Denis-Alexander Engemann Résumé : Arousal is implicated in different processes such as vigilance, alertness, attention, and affective states. Several studies have examined the correlates of this phenomenon, both at the central and peripheral nervous system levels. However, the configuration of the coupling dynamics between neural activity and behavioural measures related to this concept remains to be unravelled. One approach to address this issue is to implement predictive modelling of subjective arousal from electroencephalography data. However, it is pos- sible that this bypassing of peripheral information to explain self-reported subjective arousal is missing out on available information to further under- stand this process. Thus, a two-step approach was proposed in the present thesis. Firstly it was assessed whether it is possible to decode electrodermal activity from EEG. Then, it has been evaluated if the predicted electrodermal activity can work as a proxy measure and have additive effects ! in the sub- jective arousal prediction task. This way, the first objective was achieved: it was possible to implement a regression pipeline that effectively predicts the electrodermal activity at the event-level with the data obtained from EEG recording. However, no additive effects were found in the explanation of affective arousal by the predicted electrodermal activity signal. Importantly, a programmatically-based methodology for predictive arousal modelling was established in the present thesis. This would allow extending this to other peripheral measures and other databases to achieve a better understanding of the arousal phenomenon. Mots-clés : Arousal decoding, Predicting modelling, EEG, EDA 78 p., 2021 Laboratoire de stage : Laboratoire Parietal, Inria Saclay Spécialité : Intelligence artificielle Cote pour l’emprunt : Master 1948 ______________________ • DANAN Flora Segmentation différée d’évènements complexes en mémoire épisodique Sous la direction de Fabrice BERNA Résumé : La recherche présente s’inscrit dans un projet appelé ‘‘Bridging the gap between episodic and autobiographical memory’’. Ce projet tend à employer des dispositifs technologiques modernes comme la réalité virtuelle ou les caméras portatives afin d’étendre les données sur la mémoire épisodique à un cadre écologique qui tend à se rapprocher de la mémoire autobiographique. Il vise à mieux comprendre les modalités d’encodage et le traitement cognitif d’évènements dynamiques, complexes, multisensoriels. Dans le cadre de la théorie de la segmentation d’évènements (Zacks et al., 2007), il a été montré que mettre en exergue le processus de découpage mental des séquences d’actions pourrait améliorer l’encodage en mémoire. Afin d’être en mesure d’évaluer la segmentation des évènements personnellement vécus dans des études à venir, nous avons souhaité savoir si l’amélioration de la mémoire liée à une segmentation explicite des évènements serait toujours observable en introduisant un délai entre les phases d’encodage, de segmentation et de récupération. Par ailleurs, nous avons souhaité tester l’effet de lâ! �™implication du sujet dans la tâche en manipulant la perspective (1ère personne vs 3e personne) adoptée lors du visionnage des vidéos et de la segmentation d’évènements. Nos analyses réalisées sur un échantillon de faible taille n’ont pas montré de différence entre les groupes segmentation et contrôle, mais il a été constaté que les vidéos visualisées en perspective 1ère personne bénéficiaient d’un meilleur rappel que les vidéos en 3ème personne. Des analyses secondaires ont été réalisées afin d’identifier des variables pertinentes pour la conception expérimentale de la suite du projet, qui évaluera les mécanismes de la segmentation d’événements encodés en vie réelle. La suite de ce travail consistera à mieux comprendre les altérations d’encodage et de segmentation d’évènements dynamiques chez les personnes avec schizophrénie. Mots-clés : segmentation d’évènements, encodage, perspective, vidéos, mémoire épisodique, mémoire autobiographique. 68 p., 2021 Laboratoire de stage : Neuropsychologie Cognitive et Physiopathologie de la Schizophrénie Hôpitaux universitaires de Strasbourg Spécialité : Psychologie Cote pour l’emprunt : Master 1907 ______________________ • DRACHEVA Alina Mutual exclusivity as bayesian inference: In language and beyond Sous la direction de Benjamin Spector et Emmanuel Chemla Résumé : The present thesis explores the rational Bayesian reasoning account of the mutual exl- cusivity (ME) inference behind the well-documented ME effect. In this work, I present a formal model of a simple ME task, in which the ME effect is derived through a mere probabilistic inference, even in the presence of a small bias favouring the lexica that con- tain synonymy. In the second part of the thesis, I present an experimental paradigm, which would allow for a direct comparison of the size of the ME effect in two isomor- phic contexts – one communicative and linguistic, and another one purely mechanical, and, as such, make it possible to quantitatively measure the specifically linguistic and communicative part of the ME effect. Mots-clés : 33 p., 2021 Laboratoire de stage : Institut Jean Nicod, CNRS-ENS-EHESS Spécialité : Langage Cote pour l’emprunt : Master 1947 ______________________ • DROBI GUSEVA Marina Revue de littérature pour enrichir les méthodes d'intervention auprès des enfants socialement vulnérables et de leurs familles à partir de l'état des connaissances en science cognitive Sous la direction d’Alejandrina Cristia et Roselyne Masson Résumé : Ce mémoire de Master est issu d’un stage de recherche appliquée au sein du Conseil départemental de la Seine-Saint-Denis. Son objectif est de proposer des pistes pour enrichir les actions de soutien à la parentalité menées par le département. Le corpus de connaissances de plus en plus étayé atteste de l’importance de la régulation émotionnelle dans le développement des enfants. Il s’agit d’une compétence qui se développe progressivement grâce aux interactions avec les adultes et certaines pratiques parentales soutiennent l’acquisition de stratégies de régulation émotionnelles adaptatives. Ce constat a conduit à l’émergence d’interventions visant à soutenir les parents dans l’acquisition et la mise en œuvre de ces pratiques. Ce travail de revue systématique de la portée a pour but d’identifier si ces programmes peuvent pallier les impacts développementaux des situations de vulnérabilit? que vivent certains enfants, en restituer les conditions de déploiement, identifier comment ces connaissances peuvent enrichir les pratiques des professionnels et identifier les axes de recherche future. Cette revue a été réalisée à partir de 13 études incluses selon les critères suivants : publiées de 2005 à 2020 et revues par les pairs ; portant sur l’évaluation d’interventions parentales visant le renforcement de la régulation émotionnelle des enfants ; en français ou en anglais ; études expérimentales ou quasi-expérimentales en France, dans les pays limitrophes ou dans les pays anglo-saxons ; présence de mesures du développement de l’enfant ; absence de ciblage d’un public très restreint. A partir d’une cartographie exhaustive des connaissances, cette revue permet de conclure à la pertinence des programmes étudiés pour infléchir la trajectoire développementale des enfants. Leurs cadres théorique et opérationnel sont compatibles avec les modes opératoires des services du département même si un travail d’adaptation est nécessaire et le corpus de connaissances doit être approfondi. Mots-clés : soutien à la parentalité, programme, émotions, développement émotionnel, régulation émotionnelle, accompagnement émotionnel, socialisation des émotions 111 p., 2021 Laboratoire de stage : Direction Enfance Famille du conseil départemental de la Seine-Saint-Denis Spécialité : Ingénierie cognitive Cote pour l’emprunt : Master 1946 ______________________ • DUSSARD Claire Effets de la transparence tâche-feedback sur le neurofeedback beta en imagerie motrice Sous la direction de Charlotte Jacquemot Résumé : Ce rapport présente les résultats de nos travaux visant à étudier l’effet de la transparence tâche - feedback sur la performance en neurofeedback ? avec imagination motrice. Trois conditions de feedback unimodales et multimodales de transparence croissantes ont été testées dans un design intra-sujet avec 22 participants. Le feedback délivré au sujet était lié à leur niveau de désynchronisation dans la bande ? au sens large (8-30Hz), calculé au niveau de l’électrode C3, contralatérale au mouvement imaginé de la main droite. Nous obtenons une désynchronisation plus forte lorsque le feedback visuel est transparent (main virtuelle reproduisant le type de mouvement imaginé versus pendule oscillant plus ou moins selon le niveau de désynchronisation). En revanche, il n’y a pas de bénéfice du feedback multimodal (main virtuelle + vibration intermittente sur le dos de la main du sujet induisant une illusion de mouvement). Ces résultats sont in! terprétés en lien avec la transparence du feedback et en termes d’agentivité perçue par le sujet, qui était plus forte dans la condition de feedback visuel transparent (main virtuelle). Mots-clés : Imagination motrice, neurofeedback, agentivité, activité Beta 59 p., 2021 Laboratoire de stage : Institut du cerveau et de la moëlle épinière Paris Spécialité : Neurosciences Cote pour l’emprunt : Master 1908 ______________________ • EL OTMANI Elias A computational approach to nitoctinic modulation of working memory and related pathologies Sous la direction de Boris Gutkin Résumé : Nicotinic neuromodulation of the prefrontal cortex has long been known to be key to cognitive processes such as attention and working memory, and more recent data unravels the distinct roles of different nicotinic acetylcholine receptor subtypes as well as those of tonic and phasic modes of cholinergic release. However, how these features underpin network activity to determine cognition is not known, and a mechanistic explanation of nicotinic pathophysiology is therefore lacking. In the present work we review theories of nicotinic modulation of prefrontal persistent spiking processes, while testing them on a computational model of supragranular microcircuitry. Our simulations match the majority of the empirical findings we proposed to reproduce, and further suggest opposite and complementary roles of tonic and phasic modes of cholinergic release on the stability of WM representations. Moreover, our framework accounts for some of the clinical phenotypes of nicotini! c alterations, and provides a mechanistic basis for the self-medication hypothesis of heavy smoking in schizophrenia. Mots-clés : 44 p., 2021 Laboratoire de stage : Laboratoire de neurosciences cognitives et computationelles, Inserm-ENS Spécialité : Neurosciences Cote pour l’emprunt : Master 1945 ______________________ • FARKAS Bence Csaba Understanding individual differences in reward-guided learning as an efficient adaptation to task uncertainty and early life adversity: a pre-registered in a large online sample Sous la direction de Pierre O. Jacquet et Valentin Wyart Résumé : Humans have been shown to be capable of adjusting their learning rates in response to expected and unexpected uncertainty in the environment. Moreover, their inferences have also been shown to be subject to computational noise, driving exploratory decisions. However, it has not been investigated how these two parameters are set in a coordinated manner during reward guided learning. To fill this gap in our understanding, we have recruited 175 human subjects to take part in a decision-making task administered online, in which they had to take reward maximizing actions in an environment that was characterized by both expected and unexpected uncertainty. Their behaviour was modelled with a noisy Kalman filter to estimate individuals’ perceived uncertainty, learning rates and computational noise levels. Our results indicate that learning rates and computational noise levels covary in a systematic manner, and that their covariation is guided by the perceived expected ! and unexpected uncertainty of subjects. Moreover, there is evidence for both state and trait like sources of inter-individual variability. This work also serves as a pilot study for another preregistered study, which will go beyond the current work by trying to uncover the origins of such inter-individual variability in the early life experiences of people. Mots-clés : Uncertainty; Learning rate; Computational noise; Behavioural modelling; Reinforcement learning; Early life adversity 66 p., 2021 Laboratoire de stage : Laboratoire de neurosciences cognitives et computationnelles, Inserm-ENS Spécialité : Neurosciences Cote pour l’emprunt : Master 1944 ______________________ • FERMO Aurélien Inverse Reinforcement Learning for Shift Divided Attention Modeling Sous la direction de Caroline Chanel et Bertille Somon Résumé : How to infer, from electrophysiological data, the strategies followed by an agent in order to treat multimodal sensory information at the same time? To answer this question, we suggere to rely on the Markov Decision Process framework and to treat attentional foci as the different decisions of an artificial brain. The first challenge is to be able to infer attentional decisions from the sole observations of effective EEG connectivity. When an agent switches its attention from a visual stream of stimuli to an auditory one, we want that our model be able to map the observed connectivity pattern to that precise attentional decision. Moreover we want to model the way and the reason why, at a given moment, a brain’s agent decided for instance to keep focusing on auditory stimuli for a while. This is the second challenge: we need a new and original method of infering attentional strategies over time. Here we suggest to apply Inverse Reinforcement Learning and show how ! to process and model our EEG data in order to come to that final point. To our knowledge, this has never been done before. Our pipeline requires to split the problem in half with a different experiment for each part. The first experimental part serves at controlling which connectivity patterns is associated with which attentional decision – it gives us, in a sense, our training set. The second experiment would relax the attentional constraints and allow participant to attend freely to either stream of stimuli separately, at the same time, sequentially, etc – it would be our testing set. Only the first experiment has been launched so far but we still want to present the whole pipeline. Yet, for now, we showed that the classical EEG analysis relying on steady-states visually evoked potential and auditory steady- states responses are not useful for identifying attentional focus and dynamics. So instead we identified the relevant clusters of electrods that showed difference! in signal between attentional decisions. Then, we applied a Multivariate Autoregressive model on the different clusters of electrods and computed a measure of directed connectivity. The resulting connectivity matrices, integrated in different frequency ranges, were then fed into a Convolutional Neural Network that was supposed to learn and classify correctly the connectivity patterns. Unfortunately we did not manage to make it learn. We discuss at the end possible improvements. Mots-clés : intermodal attention, EEG connectivity, MVAR modelling, Markov Decision Process, Inverse Reinforcement Learning 36 p., 2021 Laboratoire de stage : ISAE-Supaero, Toulouse Spécialité : Psychologie Cote pour l’emprunt : Master 1943 ______________________ • GALLEZOT Charlotte Expression of emotions through spoken language in Huntington’s Disease Sous la direction de’Anne-Catherine Bachoud-Levi, Emmanuel Dupoux Résumé : Huntington’s disease (HD) is a rare neurodegenerative disease tha t is characterised by progressive motor, cognitive and psychiatric disorders. Patients with HD are impaired in emotion recognition through face, voice and body posture. Less is known about their ability to express emotions. In particular, there are no results on emotion expression through spoken language. The aim of our study was to investigate emotion expression through voice and spoken language in HD. Methods used in previous studies to assess emotion expression in HD or to study emotional prosody were not relevant for our investigation. To investigate emotion expression in HD we used emotion classifiers. We found that symptomatic patients with HD are impaired in emotion expression through voice and spoken language whereas there was no evidence for a similar impairment in premanifest pa- tients. Following our results we tried to understand which underlying mechanisms could account for these impairments. Our results were consistent with the hypothesis brought by [1] of an impair- ment in motoric representation of emotions. Yet what we observed on emotion expression through language indicates that this hypothesis is not sufficient to explain emotional dysfunctions in HD. Our results bring new perspectives for future studies. First, we highlighted several questions that should be investigated as to the precise nature and chronology of HD patients’ emotional impairments. Then, if thoroughly validated, our method could be used in future studies on other diseases. Beyond these academic perspectives, our results also have clinical implications. Awareness of HD patients’ emotion expression impairments should be raised amongst patients and caregivers to improve their interactions. Then, we showed that even if it was harder to classify emotions in HD patients’ speech than in controls’, it was still possible to do so. This paves the way for the development of new methods to monitor HD at home. 31 p., 2021 Laboratoire de stage : LSCP, NPI Spécialité : Langage Cote pour l’emprunt : Master 1909 ______________________ • HADDAD Rhea Optimiser un dispositif de participation citoyenne qui vise à réduire l’impact environnemental de la filière bovine à Nantes Métropole grâce aux sciences cognitives Sous la direction de Dominique Barreau et Thibaud Griessinger Résumé : Ce projet s’inscrit dans le cadre du Projet Alimentaire Territorial de Nantes Métropole qui vise à répondre aux enjeux de sécurité alimentaire, de transition écologique, ou encore de réduction de gaspillage alimentaire. L’objectif de ce projet est d’utiliser les sciences cognitives dans le but d’optimiser une méthodologie de participation citoyenne qui réunit les acteurs de la filière bovine (de l’éleveur au consommateur en passant par les acteurs intermédiaires) afin de les accompagner dans la rédaction d’un plan d’action visant à réduire l’impact environnemental de leurs pratiques. Ce dispositif participatif permet de prendre en compte l’interdépendance des acteurs de la filière dans l’appréhension de la complexité de la problématique en se reposant sur les expertises et points de vue des acteurs qui seront mis en situation et outillés de connaissances pour qu'ils puissent appréhender la complexité de la problématique ! et des enjeux environnementaux, et comprendre leur marge d’action et les points de vue des autres acteur pour ensemble dégager des pistes d'action. A partir de l’identification d’obstacles cognitifs majeurs à la réflexion et à la prise de décision collective nécessaire à la mise en place d'un plan d'action ambitieux, nous avons développé une méthodologie d’ateliers participatifs permettant de tester et comparer différentes stratégie pour les lever et faciliter le travail en groupe. Une méthodologie de mesure d’impact a été développée pour évaluer le succès de la démarche (qualité du plan d’action et niveau d’engagement des participants à changer de pratiques), et estimer la pertinence absolue et relative des leviers mis en place pour déjouer les obstacles cognitifs identifiés. Mots-clés : participation citoyenne, prise de décision en groupe, polarisation de groupe, pensée de groupe, paresse sociale, leviers d’action. 109 p., 2021 Laboratoire de stage : Inserm - Acte/Lab Spécialité : Cognition sociale Cote pour l’emprunt : Master 1910 ______________________ • HERRERA-MASUREL Alicia Nuclear Energy and CO2 Emissions : bridging the consensus gap on the contribution of nuclear energy to climate change Sous la direction d’Hugo Mercier Résumé : Even if the topic of the use of nuclear energy is controversial today among experts, there is a consensus on the fact that it emits a very low amount of greenhouse gas and therefore that it represents a possible solution in the fight against climate change. Yet, the fake idea according to which nuclear energy emits large amounts of CO2 is world widely spread among non-experts, and this belief might lead to wrongfully discarding a technology potentially useful in the fight against climate change. We experimentally analysed the origins of this belief, comparing two theories that could explain its existence and its spread : it could either be considered as an expression of a general negative opinion on nuclear energy stemming from the general dread it provokes (bottom up approach), or it could be considered as a specific belief held from a direct source of wrongful information (top down approach). We elaborated an experiment on the intuitive association of nuclear en! ergy to dangerous and harmless fake products to test whether it was considered as more harmful and negative than other energy sources or not. Our results showed that there existed a general negative opinion on nuclear energy, concurring with the bottom up approach that acknowledged the existence and the spread of this belief. We then tested the efficiency of exposing non-experts to arguments on the low level of nuclear energy CO2 emissions in order to change their opinion on this specific topic, and analysed whether it also impacted their belief that nuclear energy contributed to climate change. Our results showed that the belief according to which nuclear energy emitted CO2 was not deeply entrenched and easy to dismantle, and that argumentation on CO2 emissions had a positive impact on the perception that nuclear energy didn't contribute to climate change. But compared to other topics, such as nuclear radiations, nuclear benefits in terms of energy supply and economics, th! e specific topics of climate change and CO2 weren't effective enough to substantially change participants’ general opinion on nuclear energy. Mots-clés : Nuclear energy, Greenhouse Gas, Argumentation, Climate change, Fake ideas, Consensus Gap 41 p., 2021 Laboratoire de stage : Institut Jean Nicod, Paris, équipe Cognition sociale - EDF Lab - ICAME Spécialité : Cognition sociale Cote pour l’emprunt : Master 1911 ______________________ • KEREBEL Adrien Étude des interactions entre mind-wandering et métacognition Sous la direction de Jérôme Sackur Résumé : L’étude de ce qu’il est admis d’appeler la “rêverie éveillée” (mind-wandering), montre qu’une même personne se trouve successivement dans des états mentaux variés : son attention peut se détacher de la tâche en cours, et son degré de concentration fluctuer (Smallwood & Schooler, 2015). La nature de ces fluctuations a été maintenant bien étudiée, ainsi que leurs conséquences sur le niveau de performance global des individus. En revanche, l’étude des interactions entre ce niveau “meso” et le niveau “micro” des mécanismes cognitifs élémentaires est encore à l’état embryonnaire. Dans ce projet, nous proposons une première approche de cette question, en nous concentrant sur la métacognition. En sélectionnant parmi les facultés métacognitives la détection d’erreur, nous testons d’une part si les fluctuations de la concentration affectent ces capacités par- delà leur impact bien connu sur la performance elle-même (direction meso Æ micro); et réciproquement, nous étudions la possibilité que la détection d’erreur ait un rôle causal dans la régulation de l’état mental (direction micro Æ meso). Pour cela, pour utilisons une tâche d’attention soutenue dans laquelle les sujets ont la possibilité de détecter leurs erreurs (Error Awareness Task, Hester et al. 2005). Des « sondes de pensées » sont régulièrement présentées aux participants pour qu’ils rapportent leurs états mentaux. Dans un premier temps, des modèles de régression sont utilisés pour prédire d’une part les erreurs et les détection d’erreurs à partir de l’état mental, et d’autre part les états mentaux à partir de la performance et de la capacité à détecter ses erreurs. Les résultats montrent un lien local potentiel entre détection d’erreur et niveau de concentration subjectif. Dans un second temps, nous proposons une méthode d’analyse causale qui suggère que le taux d’erreurs, mais pas le taux de détection d’erreurs, a une influence unilatérale sur le niveau de concentration à grande échelle. Mots-clés : mind-wandering, rêverie éveillée, états mentaux, métacognition, détection d’erreur, causalite 30 p., 2021 Laboratoire de stage : LSCP, Paris Spécialité : Langage Cote pour l’emprunt : Master 1912 ______________________ • KONUK Can Explaining the successes of reasoning without Intellectualist postulates Sous la direction de Salvador Mascarenhas Résumé : Most research on reasoning has relied on an ‘Intellectualist’ postulate: reasoning is after truth. Logical mistakes are bugs of reasoning which Intellectualist theories should build additional hypotheses to account for. Recently, the Interactionist theory of reasoning (Mercier Sperber, 2011, 2017) proposed a different approach. Reasoning is not for tracking truth but for convincing others with arguments. This converse approach comes with a converse problem : how can humans consistently agree on logical norms and show a preference for logically valid arguments? This problem is hard to fix within the Interactionist framework without reintroducing Intellectualist assumptions. I argue however that a solution can be offered by cashing out the Interactionist theory with a lower-level theory of reasoning, the Erotetic theory of reasoning (Koralus & Mascarenhas, 2013). I prove that certain structural properties of the Erotetic system make it so that more logical argum! ents systematically rely on less informative premises as a coexten- sive feature. This allows an evaluation function that tracks the informativeness of arguments, in the style of the Interactionist theory’s ‘Epistemic vigilance’, to ef- fectively favor logical reasoning in Interactions without being tailored to classical logic. This proof will have an important theoretical implication: we can explain the normative power of classical logic on human reasoning without assuming that the norm of classical logic is implemented anywhere in the human computational system. Mots-clés : Psychology of reasoning, Interactionist theory of reasoning, Erotetic theory of reasoning, Classical logic. 40 p., 2021 Laboratoire de stage : Institut Jean Nicod, Paris, équipe Linguae Spécialité : Philosophie Cote pour l’emprunt : Master 1913 ______________________ • KOSHEVOY Alexey Contact-induced changes in colexification patterns Sous la direction de Thierry Poibeau, Alexandre François et Siva Kalyan Résumé : Languages differ significantly in ways in which they partition their semantic space. Although researchers study semantic partitioning in many different ways, the idea of colexifications (François (2008)) offers a way to explore cross-linguistic patterns of polysemy. For instance, languages such as Russian (< Slavic) or Archi (< Nakh-Daghestanian) colexify the concepts FOOD and MEAL1. The same pattern is shared across 136 other languages from different linguistic families, according to the CLICS database (Rzymski et al., 2020). On the other hand, the colexification of the concepts BREAD and FLOUR The research question that this study tries to answer is the following: what factors guide the distribution of colexification patterns across languages? Although similar research questions are addressed in the literature, the results seem to be very polarized. For instance, while studying the distribution of colexification patterns related to celestial objects and landsca! pe, Youn et al. (2016) have not found any effects of factors such as areal or phylogenetic prox- imity. On the other hand, Jackson et al. (2019) claim that the colexification patterns of emotion-related concepts are distributed according to areal proximity, although this factor alone does not fully account for their distribution. Thompson et al. (2020) have examined the distribution of the colexification patterns and have found no effect of areal, cultural, or genetic proximity. Although those studies have all used different datasets and metrics to measure the similarities in colexification repertoires, they reflect the long-going debate between the so- called universalist and relativist views on the lexicon. The first view’s central assumption is that semantic partitioning should be primarily universal as it reflects the cognitive constraints shared across the world’s population. The second assuming that the semantic partitioning should be mostly language-specific, as ! it is shaped by extra-linguistic factors, such as areal or phylogenetic proximity. The central hypothesis is that the language contact drives the distribution of the colexification patterns, contrary to other factors, such as phylogenetic proximity, cultural similarity, or con- ceptual space’s universal structure. Data from 105 languages from the NorthEuraLex database (Dellert et al., 2020) is used to test this hypothesis. The Weighted Minimum Colexification Distance (WMCD) is introduced to measure the distance between languages in relation to shared colexification patterns. WMCD is computed as follows. First, the inverse3 probability of each colexification in the database is computed. Given the set of shared colexifications between two languages, each is weighted according to its inverse probability. Then, the sum of weights is taken and divided by the sum of the total number of colexifications in two lan- guages. The main goal of introducing this metric is to quantify the rarely shared innovations in the colexification repertoire, which is not possible! with other set similarity metrics such as the Jaccard index used in similar studies (Youn et al., 2016; Thompson et al., 2020). WMCD offers robust results that are stable even if the number of concepts is relatively low4. The 105 language varieties from 21 language families are combined in 5460 pairs. For each pair, the geographic distance (median = 3351.7, sd = 2284.049) and WMCD (median = 0.09159214, sd = 0.02415513) are computed. Whether the two languages belong to the same family and sub-family is also inferred from the initial database. Following (Yamauchi and Murawaki, 2016), the possible contact situations are inferred through a threshold distance in a neighbor graph of languages, where the threshold value is set at 1000 km. As the distribution of WMCD does not follow the normal distribution, a non-parametric Mann-Whitney U test was used to compare WMCD values relative to different groupings. As the difference was found significant with regards to both contact and phylogenetic grouping, additional testing was performed. Additional testing is done using the Gradient Boosting Regression, as this ML model provides good accuracy on different datasets and is highly interpretable. This testing indicated that nor contact nor distance can be used as predictors for WMCD. However, the population difference and phylogeny have been found to be important predictors. Additionally! , the Neighbour Pressure Metric from Nikolaev (2019) is used to infer whether the exposure to a particular colexification pattern can be used to predict its presence in a given language. The NPM metric was found to be a significant predictor for the presence or absence of particular patterns when combined with both phylogenetic attribution and the probability of occurrence. This result supports the hypothesis of contact-induced distribution of colexifications. 1https://clics.clld.org/edges/256-1526 2https://clics.clld.org/edges/1368-1594 3Probability subtracted from one. 4We found that the lowest number of concepts to produce consistent results is 50. 80 p., 2021 Laboratoire de stage : Lattice, Paris Spécialité : Langage Cote pour l’emprunt : Master 1914 ______________________ • LAMRANI ALAOUI El Ghali Applying behavioral science to online consumer protection Sous la direction de Mariam Chammat et Coralie Chevallier Résumé : Consumers are increasingly vulnerable to online fraud. Many causes explain this observation of the Direction Générale de la Concurrence, de la Consommation et de la Répression des fraudes (DGCCRF). First, consumers are bounded by their limited attention in an environment that is becoming increasingly vast and complex. Second, consumers have emotions that can be manipulated to incite a target behavior and vendors are more aware of that. Third, there is an asymmetry of information between vendors and consumers that make the latter more vulnerable to ill-intentioned practices. To solve this problem, we can either change the environment or change the decision maker, in this case, the consumer. A tripartite team with people from the Direction Interministérielle de la Transformation Publique (DITP), the Behavioral Insights Team (BIT) and the DGCCRF decided on a preventive approach that empowers consumers to make better decisions. This approach involves creating a te! achable moment and training consumers. An experimentation is currently unfolding in the field to confirm the effectiveness of this approach. We do not have concluding data at this point but we hope it will prove useful and become a new tool in the fight against online fraud. 70 p., 2021 Laboratoire de stage : Institut Jean Nicod, équipe Cognition sociale, Paris - Direction interministérielle de la transformation publique Spécialité : Cognition sociale Cote pour l’emprunt : Master 1915 ______________________ • LANNELONGUE Elisa Bayesian modeling of temporal predictions Sous la direction de Sophie Herbst 72 p., 2021 Laboratoire de stage : CEA, DRF/Joliot, INSERM, Cognitive Neuroimaging Unit Spécialité : Neurosciences Cote pour l’emprunt : Master 1916 ______________________ • LEPREVOST Florian Functional Ultrasound imaging of ferret auditory cortex after sound exposure reveals spontaneous pattern reactivations modulated by novelty and relevance Sous la direction d’Yves Boubenec Résumé : Spontaneous neural activity is widely thought to be more than noise, but its nature, content or function is still largely unknown. Literature from memory research suggests that subsequent reactivation of sensory experience underlies memory consolidation, and such reactivations could thus be part of spontaneous activity. However, reactivations have been mostly observed with a specific neuronal population (hippocampal place cells), but almost never in the cortex. Furthermore, if we largely know what predicts memory performance, we don’t know what predicts reactivations. Here we imaged a ferret’s auditory cortex during and after sound exposure, to test for the presence of reactivation of recent sensory experience in spontaneous activity. We used a large field imaging method (functional ultrasound) well suited to look at pattern similarity between cortical activity during and after sound exposure. We looked at the similarity of sound exposure periods with subsequent resting state periods, and as a control we contrasted it with the similarity with pre-exposure rest periods. The similarity with the post-exposure period was significantly higher, suggesting the presence of reactivations. The effect was more important the first time sounds were heard, as well as for more ecologically relevant sounds (ferret vocalizations). These results tend to confirm that reactivations of sensory experience are indeed present in spontaneous cortical activity. Moreover, the quantity of reactivation is increased by the novelty and the relevance of the experience. We thus continue to better understand spontaneous activity, and how it is influenced by recent sensory experience. Finally, this provides an original argument for the presence of cortical reactivations, and shows that they are influenced by the same factors as memory. 38 p., 2021 Laboratoire de stage : Laboratoire des systèmes perceptifs, ENS-CNRS, Paris Cote pour l’emprunt : Master 1917 ______________________ • LORIN Louis-Marie R&D: Détection d’outliers et création de builtins Sous la direction d’Emmanuel Chemla Résumé : L’IA de Golem.ai ne traite pas toutes les données textuelles en entrée pour faire son analyse linguistique, mais l’on sait que parmi ces rebuts, il existe des outliers sémantiquement significatifs. Notre POC suggère qu’une analyse statistique via une version modifiée de tf-Idf pourrait permettre d’identifier ces outliers de manière automatique. Par ailleurs, l’IA de Golem.ai n’était pas configurée pour détecter des builtins, des éléments textuels structurés que l’on retrouve quel que soient les besoins clients. En particulier, j’ai dû configurer l’IA pour pouvoir détecter les adresses postales et les dates. Mots-clés : Intelligence artificielle symbolique, NLU, détection d’outliers, sémantique, Golem.ai, tf-idf 37 p., 2021 Laboratoire de stage : Entreprise : Golem.ai Spécialité : Intelligence artificielle Cote pour l’emprunt : Master 1918 ______________________ • MARANT Mariette Conception d’un atelier de rééducation clinique en réalité virtuelle : côté thérapeute et côté patient Sous la direction de Valérian Chambon - Marie Lourenço Résumé : La réalité virtuelle tend maintenant depuis quelques années à s’utiliser dans le domaine de la santé, en montrant déjà de nombreux bénéfices et en répondant à des besoins jusque-là restés insatisfaits. Des outils innovants voient ainsi le jour, tels que le logiciel Good Cells, développé par l’entreprise lilloise My Cyber Royaume, pour la rééducation fonctionnelle des troubles cognitifs et moteurs. Ce logiciel regroupe des ateliers de rééducation clinique en réalité virtuelle, parmi lesquels figure “Tir sur meubles clignotants”, un atelier d’archerie sollicitant les capacités d’exploration visuo-spatiales, créé pendant ce stage. Il pourrait bien s’avérer efficace et pertinent pour le traitement de la négligence spatiale unilatérale, un syndrome complexe et hétérogène qui suscite encore beaucoup d’interrogations, notamment en ce qui concerne sa rééducation. L’objectif de ce mémoire est double : évaluer, côté pati! ent, l’effet thérapeutique de cet atelier, et tester, côté thérapeute, la pertinence des mesures logicielles choisies, dans le cas particulier du traitement de la négligence spatiale unilatérale, avec pour hypothèses que l’atelier permet une rééducation plus rapide et plus durable et que celle-ci transparaît dans les mesures logicielles. Pour le vérifier, je propose un protocole en trois groupes indépendants (réalité virtuelle avec l’atelier, réalité virtuelle sans l’atelier et prise en charge classique), avec bilans neuropsychologiques en pré/post l’intervention et follow-up pour évaluer l’effet thérapeutique de l’atelier et courbes de progression des mesures du logiciel pour tester la pertinence de celles-ci. Le contexte sanitaire n’ayant pas permis son implémentation, je discute ce protocole à la lumière de l’écart entre l’idéal de la recherche et la réalité du terrain, à l’origine de compromis, et je développe les aspect! s clés de la recherche et du développement d’outils en réalité virtuelle pour la santé que ce projet a permis de mettre en évidence. Mots-clés : Réalité virtuelle - rééducation - traitement - négligence spatiale unilatérale - héminégligence 142 p., 2021 Laboratoire de stage : Institut Jean Nicod, Paris - My Cyber Royaume Spécialité : Ingénierie cognitive Cote pour l’emprunt : Master 1919 ______________________ • MC DOUGALL -VIGIER Louis Modelling the cerebellum with a reinforcement learning algorithm Sous la direction d’ Alex Cayco-Gajic Résumé : The cerebellum is classically thought of as contributing to motor learning by implementing a supervised learning rule, whereby the cerebellar cortical circuit integrates feedback errors with sensory expectations to refine movements. However, a growing body of research suggests that the workings of the cerebellum might involve an unexpected variety of learning rules, and that this brain structure partakes in non-motor processes, including cognitive ones. In particular, recent studies highlighted the cerebellum's role in reinforcement learning, i.e. learning through rewards or punishments. Electrophysiological and imaging studies in mice revealed that, in conditioning tasks involving rewards, most components of the cerebellar cortical microcircuit encode reward-related information such as prediction errors and reward expectations. Yet, how those reward-related signals are precisely integrated by the cerebellum is unknown. In this project, we set out to account for t! he integration of reward-related information in the cerebellum with an algorithm from the field of reinforcement learning. We formulated the novel hypothesis that the cerebellar cortical microcircuit implements a reinforcement learning algorithm known as semi-gradient SARSA, based on current knowledge about how different neuronal types encode reward, sensory and motor information. To test this hypothesis, we first built a model of the cerebellar cortical microcircuit, in the form of an agent learning through a semi-gradient SARSA algorithm. We tested the learning capabilities of that agent in two benchmark reinforcement learning tasks that involve goal-directed motor control : the inverted pendulum, or "cartpole" task and the "lunar lander" task. We show that our model can learn to achieve an acceptable performance in those two tasks, suggesting that semi-gradient SARSA could account for reward-related learning in the cerebellum. Finally, we discuss caveats and potential fu! ture steps that would strengthen our conclusion. Mots-clés : reinforcement learning ; cerebellum ; semi-gradient SARSA ; motor control 38 p., 2021 Laboratoire de stage : LNC2-Paris Spécialité : Neurosciences Cote pour l’emprunt : Master 1920 ______________________ • MICHEL Léa The mediating role of epigenetic age in the relationship between early-life stress and brain age acceleration Sous la direction d’Esther Walton - Valérian Chambon Résumé : Early-life stress is defined as the experience of stressful or adverse conditions during a child’s development. Its experience impacts brain maturation in particular through its structure and its epigenome. However, the pathways from stressful events to brain alterations are still in debate. Our project aimed to study the direct relationships between early-life stress during the development (prenatal, childhood and adolescent periods) and brain alterations and their indirect association through changes in the epigenome. We chose to use epigenetic age as a measure of epigenetic change and brain age as a measure of brain alterations since ageing is also a measure affected by early-life stress. We expected to find positive correlations between early-life stress and brain age acceleration directly, and indirectly through epigenetic age acceleration, meaning that more stress during the development would lead to an accelerated brain age and epigenetic age. These variables were computed in three structural equation models, one for each developmental period. The prenatal and childhood models did not show good fit to the data, but the adolescent model did. We also run a general model including the three periods. Stress during pregnancy was positively correlated with brain age acceleration, stress during childhood was negatively correlated with epigenetic age acceleration and stress during adolescence was negatively correlated with brain age acceleration. These results were contrary to our expectations, but it might be due to our small sample compared to the numbers of parameters used in our models, with that limitation it is difficult to draw any conclusion. Future studies could use other cohorts to increase the sample size and improve the model by taking into account the mental health issues in adolescence and adulthood. Indeed, mental disorders are a major consequence of early-life stress, this risk factor is thought to be due to alterations in neurocognitive systems and the epigenome after early-life stress. Mots-clés : early-life stress, epigenetic age, brain age, development, structural equation model 53 p., 2021 Laboratoire de stage : Addiction and Mental Health Group, Department of Psychology, University of Bath, UK Spécialité : Psychologie Cote pour l’emprunt : Master 1921 ______________________ • MISIEK Thomas Behavorial flexibility of different coordination strategies on two reinforcement learning tasks Sous la direction de Khamassi Mehdi Résumé : Mammals display both adaptive reward seeking and punishment avoiding behaviours, which help them to successfully interact with their environment. These behaviours have been proposed to emerge not from a single unitary controller but rather from parallely processing systems. On the one hand, stimulus-response behaviors, subserved by dopamine-based learning in the striatum, are slowly acquired through repetitive interaction with the environment. On the other hand, goal-directed behavior, subserved by the interactions between the hippocampus and the prefrontal cortex, rely on the extraction of the environment's statistics to build an internal model of the world. During the last decades, reinforcement learning has proved to be a suitable mathematical framework to model both goal-directed and habitual behaviors, which are best accounted for by respectively Model-Based (MB) and Model-Free (MF) algorithm families. Nevertheless, neither of these two models can single-hand! edly reproduce the range of experimentally observed mammals behaviours. Rather, it is interactions between both strategies through the monitoring of a coordination model that has been found to replicate the best behavioral data. Additionally, some researchers have started to question whether rodents are really able to perform mental simulation and planning with an internal model of the environment. Coincidentally, a more parsimonious algorithm, the Successor-Representation (SR), have been proposed as an alternative candidate to a MB account of goal-directed behavior. In this work, we aimed to determine whether a SR-MF coordination model from Geerts et al., (2020) was better than a MB-MF coordination model from Dollé et al., (2010) at accounting for rats' behavior in two spatial navigation tasks occuring in a water-maze (Pearce et al., 1998 ; Rodrigo et al., 2006). Additionally, this work aimed to verify a claim originating from Dollé et al. (2018), which stated that only ! their associative coordination model would be able to replicate the gradient effect of Rodrigo et al.,(2006), whereas an uncertainty-based arbitration model such as Geerts' should not. Indeed, we found that only Dolle's coordination mechanism was able to replicate Rodrigo's finding. The origin of the generalisation gradient remained unclear however, as well as its absence in Geerts results. Further work will also be needed to understand whether it was the SR or the reliability-based coordination model of Geerts that hindered the gradient phenomenon. Both Geerts and Dolle's models were able to replicate Pearce et al., (1998) results. The results suggested a better match of the MB to the experimental data, mainly due to suboptimal performances of the SR. Strikingly, it was found that the coordination model of Geerts, despite relying on a reliability-based arbitration mechanism, was able to generate spatial coordination patterns. We found that this phenomenon was intrinsic to ! any fusion coordination model. ? p., 2021 Laboratoire de stage : ISIR - Sorbonne université-CNRS Spécialité : Intelligence artificielle Cote pour l’emprunt : Master 1922 ______________________ • MUS Mathilde Cognition et acceptabilité de la taxe carbone Sous la direction de Coralie Chevallier et Hugo Mercier Résumé : Most public policy debates are centered around the efficiency of policy measures, neglecting their acceptability. However, there is an important difference between the efficiency of a policy, assessed by economists, and its level of support in the population. The case of the carbon tax provides a particularly striking example: despite its potential for curbing CO2 emissions, it encounters a strong public resistance in many countries, resistance that even prompted the Yellow Vest protest movement in France. However, researchers have demonstrated that the acceptability of carbon taxation depends heavily on its design attributes, and in particular on the allocation of generated revenues. For instance, studies repeatedly reveal a clear public preference towards earmarking carbon tax revenues for environmental purposes. Earmarking is a budgeting practice by which all or a portion of collected tax revenues is dedicated to a particular sector chosen in advance, rather th! an subjected to the typical budget procedure of revenue-pooling. However, there is no agreement on the explanation for this observed preference for green earmarking of carbon taxes. We suggest that this preference stems in part from the thematic matching between the source of revenue (an environmental tax) and the use of revenue (environmental projects). In turn, this preference for ‘matched earmarking’ would result from the application of mental accounting, the fact that people keep track of their expenses by creating mental budgets that attempt to match the origin of revenues with the domain in which they will be used. Using online experiments on large samples of the British population (Npilot = 500, pre-registered main studies on representative samples: N1 = 900, N2 = 1300, https://osf.io/5nyve/), we find firm evidence that mental accounting plays a role in shaping citizens’ preference towards green earmarking of carbon taxes. A replication study conducted on a Fre! nch representative sample (N3 = 990) confirmed the robustness of this result. Mots-clés : carbon taxation, earmarking, mental accounting, acceptability, state budget 59 p., 2021 Laboratoire de stage : Laboratoire de neurosciences cognitives et computationnelles, LNC2, Inserm - ENS - Equipe cognition sociale Spécialité : Cognition sociale Cote pour l’emprunt : Master 1923 ______________________ • NARBONA SABATE Lara Ontogeny of communication in Chimpanzees : A multimodal approach Sous la direction de Katie Slocombe Résumé : Is chimpanzee (Pan troglodytes schweinfurthii) communication homologous to that of their closely- related cousin, the Homo sapiens? To shed light on this question, the ontology of chimpanzees' multimodal communication is investigated. After coding for infant chimpanzees’ gestures, vocalisations, and facial expressions, we used mothers’ gregariousness as a proxy for social exposure to investigate whether the latter has an effect on the ontogeny of infant chimpanzees’ communicative behaviours. We found that mother gregariousness did not seem to influence infant communication. Additionally, we did exploratory work on infants’ ontogeny of multimodal communication and we found that (1) infants seem to produce more gestures and facial expressions than vocalisations, that (2) the percentage of initiations and responses produced by the infant seems to be sensitive to age, with initiations decreasing and responses increasing as they get older, that (3) 44.53% of th! e produced gestures are used sequentially with other communicative signals; and that (4) multimodal signals account for 21.26% of infants’ communication. Future work should study the dependence on age of initiation and responses, and what could be the cause of responses taking over the initiations at 10 months of age. Other possible work should assess the differences between adult and infant communicative behaviours and see whether there is a gradual or critical change from one to the other. Mots-clés : chimpanzee, communication, ontogeny, multimodal, mother gregariousness, social exposure 78 p., 2021 Laboratoire de stage : University of York Spécialité : Ethologie Cote pour l’emprunt : Master 1924 ______________________ • NASIOULAS Antonios The role of anticipated feedback in decision making under risk Sous la direction de Stefano Palminteri Résumé : A number of studies have shown notable differences between description-based decisions and decisions involving experience, a phenomenon known in the literature as the description- experience gap. We conducted a series of four online factorial experiments using the repeated binary choice task in fully-described gambles to investigate the effect of post-choice feedback on risk propensity and expected value (EV)-maximization. The availability of feedback was manipulated in a within-subjects manner. Our results revealed a clear effect of feedback on risk propensity. Risky rates in the feedback blocks are higher than in the no-feedback blocks. When subjects are informed about the availability of feedback in the upcoming block, this difference emerges from the very first trial. Yet, when subjects are deprived of this information, this dif- ference emerges later. These results suggest that the effect of feedback in risk propensity is driven largely by a dispositional cha! nge rather than experience per se. Additionally, we found that repeated choice within a block tends to decrease risk propensity. In contrast, neither of these factors had a significant effect on EV-maximization rates. Model-fitting analysis using the Prospect Theory model showed that both the parameter controlling the curvature of the util- ity function (thus, reflecting risk attitudes) and the parameter determining the distortion of the probability weights significantly increase when post-choice feedback is provided. This result suggests a relative underweighting of small probabilities when feedback is provided, a result compatible with the existing literature. Mots-clés : decision making under risk, description-experience gap, prospect theory, post-choice feedback 68 p., 2021 Laboratoire de stage : Laboratoire de Neurosciences Cognitives Computationnelles ENS- INSERM Paris Spécialité : Neurosciences Cote pour l’emprunt : Master 1925 ______________________ • OLECH Alexandre Le Awe et le Sublime Sous la direction de Jérôme Dokic Résumé : Le projet entrepris dans ce mémoire cherche à comprendre la nature de la relation entre les deux notions que sont le awe et le sublime. Si cette question a été abordée de manière périphérique dans beaucoup de travaux (philosophiques et psychologiques), il n’y a pas à ma connaissance d’ouvrage ou d’article dédié directement à cette question, en dehors de l’article de Arcangeli et al. de 2020 (Arcangeli, M., Sperduti, M., Jacquot, A., Piolino, P., & Dokic, J. (2020)). Awe and the Experience of the Sublime: A Complex Relationship. Frontiers in Psychology I). Ce court article constitue une sorte de tableau des hypothèses possibles pour comprendre la relation entre awe et sublime, mais comme le soulignent les auteurs, des travaux approfondissant la question sont souhaitables. Mon projet était ainsi de répondre à ce besoin exprimé d’un approfondissement de la question. 52 p., 2021 Laboratoire de stage : Institut Jean Nicod, CNRS-ENS Paris Spécialité : Philosophie Cote pour l’emprunt : Master 1926 ______________________ • OWCZAREK Eliott Study of striatal and sub-thalamic nuclei local field potentials during a metacognitive task performed by severe obsessional compulsive disorder patients Sous la direction de Karim N’Diaye and Luc Mallet Résumé : Obsessive Compulsive Disorder (OCD) is a mental disorder characterized by obsessions (intrusive and anxious thoughts, usually associated with pervasive doubt) and compulsion (repetitive and ritualized motor or mental acts). In severe OCD which is resistant to standard medication, an experimental therapeutic alternative is Deep Brain Stimulation (DBS). Despite its effectiveness when targeting specific brain regions in the basal ganglia, like the striatum (STR) or the subthalamic nucleus (STN), its mechanisms of action remain poorly understood. Our research project aims to analyse Local Field Potential (LFP) recordings collected in the context of DBS surgery in N=8 patients with severe resistant OCD while they perform a task manipulating decision uncertainty. The task used a version of the widely used random-dot kinetograms (RDK) stimuli, where the global motion coherence was manipulated to adjust the difficulty of discrimination left- from right-ward motion. In particular, assessing one's decision uncertainty is one facet of what is known as metacognition, and given previous reports of metacognitive deficits in OCD, we investigated whether the basal ganglia regions targeted by DBS would also be involved in the metacognitive monitoring underlying decision-making in uncertain contexts, which deficits might lead to pathological doubt. Therefore, we also collected post-decision confidence ratings from the patients during the task. Behaviourally we found that OCD patients are able to rate their confidence accordingly t! o the task difficulty but not to their own performance. LFP recordings shows that the STN specifically respond to the stimulus and the STR to the feedback presentation. Given the limited number of patients, these observations need further investigation, possibly in experimental contexts combining metacognitive monitoring (confidence judgement) and metacognitive control (checking behavior), in particular to connect with earlier electrophysiological studies in OCD patients. Mots-clés : OCD, metacognition, RDK, DBS, STN, Striatum 37 p., 2021 Laboratoire de stage : Neurophysiology of Repetitive Behaviors - Paris Brain Institute (ICM) Spécialité : neurosciences Cote pour l’emprunt : Master 1927 ______________________ • PAREAU, Marie What is the EEG signature of mind wandering in healthy subjects? Sous la direction d’Alice Guesdon et Philippe Fossati Résumé : Driving on a monotonous road while thinking about his grocery list or sitting in the subway while thinking about his future work meeting, these are some of the situations that characterize mind wandering. The work of this master's research aims to define the electroencephalographic signature of mind wandering in a group of healthy subjects. 26 subjects performed during EEG recording a Sustained Attention Response Task (SART) during which they had to respond to thought probes characterizing their thoughts at the moment T. Results indicated a decrease for mind wandering events in the ratio of frontal theta (4–7 Hz) to beta (13-30 Hz) oscillations for 18 of the 21 subjects analyzed. An absence of significant difference was found for alpha power (8-12 Hz) for occipital electrodes and theta power (4-7 Hz) for frontal electrodes. The behavioral data did not show significant results (number of errors and reaction time similar between the thought probes considered as on! task and those considered as off task). Mots-clés : Mind wandering – EEG – oscillations waves p. 39, 2021 Laboratoire de stage : Institut du cerveau et de la moëlle épinière Paris Cote pour l’emprunt : Master 1928 ______________________ • PIOT Leonardo French adults but not children rely more on consonants than on vowels in lexical processing: a familiar word recognition conflict task Sous la direction de Sandrine Van Ommen et Thierry Nazzi Résumé : In the literature, consonants have been proposed to be more important than vowels in lexical activation and access processes (i.e. C-bias). However, despite a large body of evidence in the infant and adult literature, no study have investigated this bias in the context of a familiar word recognition conflict task. The aim of the present study was to evaluate the extent to which French-speaking adults and children exhibit a C-bias in familiar word recognition with a task introducing a conflict between consonantal and vocalic information. To address this issue, we designed two audiovisual experiments evaluating the existence and strength of the C-bias during lexical access of familiar words in a group of 24-month-olds and a group of adults. Both experiments tested the recognition of a target object among two different familiar objects presented on a screen. In the control condition, the proposed auditory target corresponded to one of the objects presented. In the co! nflict condition, the target differed from one of the objects by a consonant and from the other by a vowel. In both experiments, word-recognition was found in the control condition. For the conflict trials, Experiment 1 revealed the existence of a consonant bias in adulthood, while Experiment 2 failed to reveal a processing bias in the 24-month-olds . Additionally, the strength of the C-bias in adults appeared to be marginally moderated by the syllabic position of the mispronounced consonant: consonants at word onset tended to carry more weight than consonants situated in coda position. 59 p., 2021 Laboratoire de stage : INCC Paris Descartes Spécialité : Neurosciences Cote pour l’emprunt : Master 1929 ______________________ • RAUGEL Joséphine Machine learning applied to neurofeedback Sous la direction de Valérian Chambon et Fanny Grosselin Résumé : Stress is a worldwide trouble that can lead to multiple serious diseases and pain. Most solutions currently applied for this trouble and related disorders are drug-based. Neurofeedback (NF), in addition to being a non-drug solution, has been shown to cause neuromodulation of alpha activity, a neuromarker linked to relaxation. In the long term and with a suitable feedback, NF might lead to a better ability of the subject to relax and to his/her independence from the NF tool. MyBrain Technologies, a start-up derived from the ICM, launched in 2017 their NF headset for stress regulation, Melomind. To identify early cases of NF inefficiency and to personalize the use of Melomind for each user, we employed a Machine Learning (ML) approach to profile and cluster users and sessions, based on data collected throughout a 12-session training carried out for a previous study. We could identify not only individual clusters of data but also meta-clusters grouping several individuals, and intra-individual clusters grouping a few sessions each. The multi-individual meta-clusters were correlated with the distribution of alpha power throughout sessions. Using ML methods, without any physiological or psychological predictors, we also predicted data distribution within 4 labels: relaxation experienced at a future session, performance deployed at a future session, progression of relaxation over training and progression of performance over training. We obtained satisfactory accuracies (from 60% up to 90%, considerably above chance level) from the 4th training session. We could predict performance and relaxation concerning up to 8 sessions ahead of the training for already-seen users. We could also predict performance and relaxation on future sessions for never-seen users, with accuracies of 87% and 61%, respectively. This work paves the way to a better understanding of the inefficacy problem and to a personalized support for NF training, based on ML and without necessary use of predictors. Mots-clés : Stress, neurofeedback, machine learning 96 p., 2021 Laboratoire de stage : myBrain Technologies, Paris Spécialité : Intelligence artificielle Cote pour l’emprunt : Master 1942 ______________________ • RAZAFINIMANANA Mihoby Beta and gamma oscillations in the cerebello-thalamo-cortical circuit during sensorimotor adaptation in healthy subjects: an meg studyBeta and gamma oscillations in the cerebello-thalamo-cortical circuit during sensorimotor adaptation in healthy subjects: an meg study Sous la direction de Gallea Cécile Résumé : Learning is a dynamic process that involves a fast improvement of motor perfor- mance through practice by reducing the error (error correction phase), followed by the stabilization of motor performance to reach a plateau with a reduction of performance variability (plateau phase). In particular, updating motor plans using feedback of past trials is an important process that occurs prior to movement onset and allows optimiz- ing the performance. Past works have identified involved brain networks, including the cerebellum and frontal motor areas. The cerebellum has cognitive and integrative prop- erties which remain quite unexplored in humans. In contrast, the supplementary motor area (SMA)’s properties related to motor planning and learning were largely investigated. In the present study, we aim to characterize the activity time course and the temporal dynamics in the cerebellum during the preparatory period of a motor-learning task in healthy population, in comp! arison with the known temporal dynamics of the SMA. Par- ticipants performed successive trials of a reaching task with their right dominant hand under a control condition and a learning one where visual feedback was rotated, during magnetoencephalography (MEG) recordings. We evaluated the effect of learning on the preparatory potential (onset and amplitude), and on the time-frequency changes within the cerebellum and the SMA. Based on the behavioral results, learning was separated into an error correction phase and a plateau one. MEG results showed a preparatory potential in the cerebellum whose onset was earlier during the learning compared to the control condition. The time-frequency analysis showed similar characteristics of the ?-oscillations within the cerebellum and the SMA, with an increase in synchronization during the early preparatory period. We used non-invasive repetitive transcranial mag- netic stimulation (rTMS) to modulate cerebellar outputs and measure the p! ost-stimulation effect on learning-related MEG recordings and behavioral performance. However, none of these effects yielded significant results. Mots-clés : motor learning, cerebellum, supplementary motor area, repetitive TMS, beta oscillations 73 p., 2021 Laboratoire de stage : Institut du cerveau et de la moëlle épinière Cote pour l’emprunt : Master 1930 ______________________ • REICH Zoé Managing Children’s Screen Consumption Sous la direction de Mariam Chammat - Coralie Chevallier Résumé : Children spend an increasing amount of time in front of screens, and public concern is fast-spreading (Santé Publique, 2017, étude Esteban). Despite this, there is no consensus among experts on what constitutes ‘healthy screen time and/or use’. This applied research project, conducted by the French public services, aims to help families navigate around confusing information. Therefore, its goal is not to define ‘healthy screen use’ but rather to create a tool relevant and useful for as large a population as possible. This research project addresses the screen question by viewing it as a behavioural problem related to habit reversal and formation. It was designed using a behavioural science protocol, namely by first identifying behavioural barriers and levers, confronting those insights with the environment and then creating solution prototypes. While the evaluation has not been conducted yet, the plan is to assess the chosen solution on a small sample of! participants before scaling it up if results are encouraging. The chosen solution is a web-application, also downloadable on smartphone, that helps families run a self-diagnosis to point out the screen-related behaviours they find most disturbing in their household. The app then provides personalised recommendations based on the families’ answers and creates a personalised ‘family screen chart’ to apply at home. The challenging part of this project will be to bolster engagement over a long enough period of time in order to create lasting new healthy habits. The behavioural levers used within the app will be discussed, as well as the limitations of its design. 129 p., 2021 Laboratoire de stage : Direction interministérielle de la transformation publique Spécialité : Cognition sociale Cote pour l’emprunt : Master 1931 ______________________ • ROUE Aurélie Can the Content of Visual Perception Be Emotional? A Cognitive Model for Understanding Dysmorphophobia in Anorexia Nervosa Sous la direction de Frédérique de Vignemont Résumé : Anorexia nervosa (AN) is a psychiatric disorder characterized by food restriction, weight loss and body image distortion. AN subjects see themselves larger than they are. Few cognitive models have tried to explain this bodily disorder. The ones proposed mainly focus on a central distorted body image which penetrates visual perception. I aimed in this work to propose a new and . This model is based on recent works on “emotional attention”. “Emotional attention” is a separate category of attention, beyond the classical distinction between bottom-up and top-down attention. According to this theory, amygdala evaluates the emotional valence of visual stimuli and boost their perceptual processing in early sensory neural pathways, including visual cortex. Sensory neural areas process more efficiently these emotional stimuli which appear more salient. This mechanism occurs very early and is automatic. I hypothesize that one explanation of visual body distortion i! n AN is due to emotional attention. I suggest that body parts such as “thigh, belly, hips” in AN are perceived as affectively-loaded visual stimuli. Amygdala ascribes a valence to body parts at an early stage of processing and, via feedback connections to visual areas, increases their early processing in visual cortex. Body parts appear more salient and, in my theory, larger. My hypothesis is congruent with previous works on AN. However, empirical data are still missing, and further experiments are required to confirm the validity of the new model I propose. Mots-clés : anorexia nervosa, emotional attention, amygdala, early stage of processing, visual stimuli, body distortion 54 p., 2021 Laboratoire de stage : Institut Jean Nicod- CNRS-EHESS-ENS, Paris - Fondation de santé des étudiants de France, Inserm, Paris Spécialité : Philosophie Cote pour l’emprunt : Master 1932 ______________________ • ROZE Patricia Exploration of morphological representations in computational language Sous la direction d’Emmanuel Dupoux Résumé : In this work, computational language models are evaluated on a series of lexical discrimination tasks. Both simple ngram models and more complex LSTMs are studied, in both English and French. The tasks rely on the discrimination between well-formed and ill-formed words, based on derivational morphology patterns. We show that neural networks perform better than chance when baseline ngram models remain at chance, in both languages. These results show that neural networks acquire some abstract representations about morphology. However, neural networks performance remain close to chance and the specific nature of the representations acquired remains unclear. Mots-clés : NLP, Morphology, LSTM, black-box NLP, linguistic representations, language modelling 36 p., 2021 Laboratoire de stage : Laboratoire des sciences cognitives et psycholinguistique, EHESS-ENS-CNRS-Inria Spécialité : Langage Cote pour l’emprunt : Master 1941 ______________________ • SARRÉ Annahita Cerebral mechanisms of Cued Speech perception Sous la direction de Laurent Cohen Résumé : Most alphabets are based on visual coding of phonemes and syllables. However, other visual codes were developed for use by deaf people for conveying visually the sounds of speech. In particular, cued speech was designed to facilitate lip reading by specifying pronounced syllables through a combination of lip configuration, hand position and handshape. The use of this communication system improves general language skills, and reading in particular, in a deaf community characterized by a low average level of literacy. Despite its proven efficiency, and while learning this system likely involves brain transformations comparable to those associated with reading acquisition, the mechanisms of cued speech perception remain largely unknown. The goal of this internship was thus to design and set up a study that will permit to explore the brain basis of cued speech perception and its links with reading, two coexisting visual codes for language that could both compete and support each other. This project is composed of three MRI experiments, two localizers and a decoding experiment, involving 3 groups of subjects (deaf and hearing subjects cued speech users and a group of cued speech-naive hearers). Our preliminary results permitted a first over …