Search

How human prefrontal and insular regions interact while maximizing rewards and minimizing punishments is unknown. Capitalizing on human intracranial recordings, we demonstrate that the functional specificity toward reward or punishment learning is better disentangled by interactions compared to local representations. Prefrontal and insular cortices display non-selective neural populations to reward and punishment. The non-selective responses, however, give rise to context-specific interareal interactions. We identify a reward subsystem with redundant interactions between the orbitofrontal and ventromedial prefrontal cortices, with a driving role of the latter. In addition, we find a punishment subsystem with redundant interactions between the insular and dorsolateral cortices, with a driving role of the insula. Finally, switching between reward and punishment learning is mediated by synergistic interactions between the two subsystems. These results provide a unifying explanation of distributed cortical representations and interactions supporting reward and punishment learning.

Reward prediction error (RPE) signals are crucial for reinforcement learning and decision-making as they quantify the mismatch between predicted and obtained rewards. RPE signals are encoded in the neural activity of multiple brain areas, such as midbrain dopaminergic neurons, prefrontal cortex, and striatum. However, it remains unclear how these signals are expressed through anatomically and functionally distinct subregions of the striatum. In the current study, we examined to which extent RPE signals are represented across different striatal regions. To do so, we recorded local field potentials (LFPs) in sensorimotor, associative, and limbic striatal territories of two male rhesus monkeys performing a free-choice probabilistic learning task. The trial-by-trial evolution of RPE during task performance was estimated using a reinforcement learning model fitted on monkeys' choice behavior. Overall, we found that changes in beta band oscillations (15–35 Hz), after the outcome of the animal's choice, are consistent with RPE encoding. Moreover, we provide evidence that the signals related to RPE are more strongly represented in the ventral (limbic) than dorsal (sensorimotor and associative) part of the striatum. To conclude, our results suggest a relationship between striatal beta oscillations and the evaluation of outcomes based on RPE signals and highlight a major contribution of the ventral striatum to the updating of learning processes.SIGNIFICANCE STATEMENTReward prediction error (RPE) signals are crucial for reinforcement learning and decision-making as they quantify the mismatch between predicted and obtained rewards. Current models suggest that RPE signals are encoded in the neural activity of multiple brain areas, including the midbrain dopaminergic neurons, prefrontal cortex and striatum. However, it remains elusive whether RPEs recruit anatomically and functionally distinct subregions of the striatum. Our study provides evidence that RPE-related modulations in local field potential (LFP) power are dominant in the striatum. In particular, they are stronger in the rostro-ventral rather than the caudo-dorsal striatum. Our findings contribute to a better understanding of the role of striatal territories in reward-based learning and may be relevant for neuropsychiatric and neurologic diseases that affect striatal circuits.

The reproducibility crisis in neuroimaging and in particular in the case of underpowered studies has introduced doubts on our ability to reproduce, replicate and generalize findings. As a response, we have seen the emergence of suggested guidelines and principles for neuroscientists known asGood Scientific Practicefor conducting more reliable research. Still, every study remains almost unique in its combination of analytical and statistical approaches. While it is understandable considering the diversity of designs and brain data recording, it also represents a striking point against reproducibility. Here, we propose a non-parametric permutation-based statistical framework, primarily designed for neurophysiological data, in order to perform group-level inferences on non-negative measures of information encompassing metrics from information-theory, machine-learning or measures of distances. The framework supports both fixed- and random-effect models to adapt to inter-individuals and inter-sessions variability. Using numerical simulations, we compared the accuracy in ground-truth retrieving of both group models, such as test- and cluster-wise corrections for multiple comparisons. We then reproduced and extended existing results using both spatially uniform MEG and non-uniform intracranial neurophysiological data. We showed how the framework can be used to extract stereotypical task- and behavior-related effects across the population covering scales from the local level of brain regions, inter-areal functional connectivity to measures summarizing network properties. We also present an open-source Python toolbox called Frites1that includes the proposed statistical pipeline using information-theoretic metrics such as single-trial functional connectivity estimations for the extraction of cognitive brain networks. Taken together, we believe that this framework deserves careful attention as its robustness and flexibility could be the starting point toward the uniformization of statistical approaches.Graphical abstractHighlightsGroup-level statistics for extracting neurophysiological cognitive brain networksCombining non-parametric permutations with measures of informationFixed- and random-effect models, test- and cluster-wise correctionsMulti-level inferences, from local regions to inter-areal functional connectivityA Python open-source toolbox calledFritesincludes the proposed statistical methods

Cognitive functions arise from the coordinated activity of neural populations distributed over large-scale brain networks. However, it is challenging to understand and measure how specific aspects of neural dynamics translate into operations of information processing, and, ultimately, cognitive functions. An obstacle is that simple circuit mechanisms–such as self-sustained or propagating activity and nonlinear summation of inputs–do not directly give rise to high-level functions. Nevertheless, they already implement simple transformations of the information carried by neural activity.Here, we propose that distinct neural circuit functions, such as stimulus representation, working memory, or selective attention stem from different combinations and types of low-level manipulations of information, or information processing primitives. To test this hypothesis, we combine approaches from information theory with computational simulations of canonical neural circuits involving one or more interacting brain regions that emulate well-defined cognitive functions. More specifically, we track the dynamics of information emergent from dynamic patterns of neural activity, using suitable quantitative metrics to detect where and when information is actively buffered (“active information storage”), transferred (“information transfer”) or non-linearly merged (“information modification”), as possible modes of low-level processing. We find that neuronal subsets maintaining representations in working memory or performing attention-related gain modulation are signaled by their boosted involvement in operations of active information storage or information modification, respectively.Thus, information dynamics metrics, beyond detectingwhichnetwork units participate in cognitive processing, also promise to specifyhow and whenthey do it, i.e., through which type of primitive computation, a capability that may be exploited for the parsing of actual experimental recordings.

International audience; The field of cognitive computational neuroscience addresses open questions regarding the complex relation between cognitive functions and the dynamic coordination of neural activity over large-scale and hierarchical brain networks. State-of-the-art approaches involve the characterization of brain regions and inter-areal interactions that participate in cognitive processes (Battaglia & Brovelli, 2020). More precisely, the study of cognitive brain networks underlies linking local neural activity or interactions between brain regions to experimental variables, such as sensory stimuli or behavioral responses. The relation between the brain data and external variables might take complex forms (e.g. non-linear relationships) with strong variations across brain regions and participants. Therefore, powerful measures of information are required to detect complex relations and the statistical relevance at the population level should be able to adapt to the inter subject variability.

Despite being the focus of a thriving field of research, the biological mechanisms that underlie information integration in the brain are not yet fully understood. A theory that has gained a lot of traction in recent years suggests that multi-scale integration is regulated by a hierarchy of mutually interacting neural oscillations. In particular, there is accumulating evidence that phase-amplitude coupling (PAC), a specific form of cross-frequency interaction, plays a key role in numerous cognitive processes. Current research in the field is not only hampered by the absence of a gold standard for PAC analysis, but also by the computational costs of running exhaustive computations on large and high-dimensional electrophysiological brain signals. In addition, various signal properties and analyses parameters can lead to spurious PAC. Here, we present Tensorpac, an open-source Python toolbox dedicated to PAC analysis of neurophysiological data. The advantages of Tensorpac include (1) higher computational efficiency thanks to software design that combines tensor computations and parallel computing, (2) the implementation of all most widely used PAC methods in one package, (3) the statistical analysis of PAC measures, and (4) extended PAC visualization capabilities. Tensorpac is distributed under a BSD-3-Clause license and can be launched on any operating system (Linux, OSX and Windows). It can be installed directly via pip or downloaded from Github (https://github.com/EtienneCmb/tensorpac). By making Tensorpac available, we aim to enhance the reproducibility and quality of PAC research, and provide open tools that will accelerate future method development in neuroscience.

International audience; Brain functions rely on flexible communication between microcircuits in distinct cortical regions. The mechanisms underlying flexible information routing are still, however, largely unknown. Here, we hypothesize that the emergence of a multiplicity of possible information routing patterns is due to the richness of the complex dynamics that can be supported by an underlying structural network. Analyses of circuit computational models of interacting brain areas suggest that different dynamical states associated with a given connectome mechanistically implement different information routing patterns between system's components. As a result, a fast, network-wide and self-organized reconfiguration of information routing patterns-and Functional Connectivity networks, seen as their proxy-can be achieved by inducing a transition between the available intrinsic dynamical states. We present here a survey of theoretical and modelling results, as well as of sophisticated metrics of Functional Connectivity which are compliant with the daunting task of characterizing dynamic routing from neural data. Theory: Function follows dynamics, rather than structure Neuronal activity conveys information, but which target should this information be-pushed‖ to, or which source should new information be-pulled‖ from? The problem of dynamic information routing is ubiquitous in a distributed information processing system as the brain. Brain functions in general require the control of distributed networks of interregional communication on fast timescales compliant with behavior, but incompatible with plastic modifications of connectivity tracts (Bressler & Kelso, 2001; Varela et al., 2001). This argument led to notions of connectivity based on information exchange-or more generically, an-interaction‖-between brain regions or neuronal populations, rather than based on the underlying STRUCTURAL CONNECTIVITY (SC, i.e. anatomic). An entire zoo of data-driven metrics has been introduced in the literature and this chapter will review some of them. Notwithstanding, they track simple correlation, or directed causal influence (Friston, 2011) or information transfer (Wibral et al., 2014) between time-series of activity. These

Idea of the model, specification of the model and tests, implementation of the model, tests, data analysis, analysis of results, writing-up. ‡Idea of the model, specification of the model and tests, analysis of results, writing-up. ¤Specification of the model and tests, analysis of results, writing-up. * Abstract In mammals, goal-directed and planning processes support flexible behaviour usable to face new situations or changed conditions that cannot be tackled through more efficient but rigid habitual behaviours. Within the Bayesian modelling approach of brain and behaviour, probabilistic models have been proposed to perform planning as a probabilistic inference. Recently, some models have started to face the important challenge met by this approach: grounding such processes on the computations implemented by brain spiking networks. Here we propose a model of goal-directed behaviour that has a probabilistic interpretation and is centred on a recurrent spiking neural network representing the world model. The model, building on previous proposals on spiking neurons and plasticity rules having a probabilistic interpretation, presents these novelties at the system level: (a) the world model is learnt in parallel with its use for planning, and an arbitration mechanism decides when to exploit the world-model knowledge with planning, or to explore, on the basis of an entropy-based confidence on the world model knowledge; (b) the world model is a hidden Markov model (HMM) able to simulate sequences of states and actions, thus planning selects actions through the same neural generative process used to predict states; (c) the world model learns the hidden causes of observations, and their temporal dependencies, through a biologically plausible unsupervised learning mechanism. The model is tested with a visuomotor learning task and validated by comparing its behaviour with the performance and reaction times of human participants solving the same task. The model represents a further step towards the construction of an autonomous architecture bridging goal-directed behaviour as probabilistic inference to brain-like computations. Author summary Goal-directed behaviour relies on brain processes supporting planning of actions based on the prediction of their consequences before performing them in the environment. An important computational modelling approach of these processes sees the brain as a probabilistic machine implementing goal-directed processes relying on probability distributions and operations on them. An important challenge for this approach is to explain how these distributions and operations might be grounded on the brain spiking doi: bioRxiv preprint neurons and learning processes. Here we propose a hypothesis of how this might happen by presenting a computational model of goal-directed processes based on artificial spiking neural networks. The model presents three main novelties. First, it can plan even while it is still learning the consequences of actions by deciding if planning or exploring the environment based on how confident it is on its predictions. Second, it is able to 'think' alternative possible actions, and their consequences, by relying on the low-level stochasticity of neurons. Third, it can learn to anticipate the consequences of actions in an autonomous fashion based on experience. Overall, the model represents a novel hypothesis on how goal-directed behaviour might rely on the stochastic spiking processes and plasticity mechanisms of the brain neurons.

Quantifying both the amount and content of the information transferred between neuronal populations is crucial to understand brain functions. Traditional data-driven methods based on Wiener-Granger causality quantify information transferred between neuronal signals, but do not reveal whether transmission of information refers to one specific feature of external stimuli or another. Here, we developed a new measure called Feature-specific Information Transfer (FIT), that quantifies the amount of information transferred between neuronal signals about specific stimulus features. The FIT quantifies the feature-related information carried by a receiver that was previously carried by a sender, but that was never carried by the receiver earlier. We tested the FIT on simulated data in various scenarios. We found that, unlike previous measures, FIT successfully disambiguated genuine feature-specific communication from non-feature specific communication, from external confounding inputs and synergistic interactions. Moreover, the FIT had enhanced temporal sensitivity that facilitates the estimation of the directionality of transfer and the communication delay between neuronal signals. We validated the FIT’s ability to track feature-specific information flow using neurophysiological data. In human electroencephalographic data acquired during a face detection task, the FIT demonstrated that information about the eye in face pictures flowed from the hemisphere contralateral to the eye to the ipsilateral one. In multi-unit activity recorded from thalamic nuclei and primary sensory cortices of rats during multimodal stimulation, FIT, unlike Wiener-Granger methods, credibly detected both the direction of information flow and the sensory features about which information was transmitted. In human cortical high-gamma activity recorded with magnetoencephalography during visuomotor mapping, FIT showed that visuomotor-related information flowed from superior parietal to premotor areas. Our work suggests that the FIT measure has the potential to uncover previously hidden feature-specific information transfer in neuronal recordings and to provide a better understanding of brain communication.Author summaryThe emergence of coherent percepts and behavior relies on the processing and flow of information about sensory features, such as the color or shape of an object, across different areas of the brain. To understand how computations within the brain lead to the emergence of these functions, we need to map the flow of information about each specific feature. Traditional methods, such as those based on Wiener-Granger causality, quantify whether information is transmitted from one brain area to another, but do not reveal if the information being transmitted is about a certain feature or another feature. Here, we develop a new mathematical technique for the analysis of brain activity recordings, called Feature-specific Information Transfer (FIT), that can reveal not only if any information is being transmitted across areas, but whether or not such transmitted information is about a certain sensory feature. We validate the method with both simulated and real neuronal data, showing its power in detecting the presence of feature-specific information transmission, as well as the timing and directionality of this transfer. This work provides a tool of high potential significance to map sensory information processing in the brain.

In mammals, goal-directed and planning processes support flexible behaviour used to face new situations that cannot be tackled through more efficient but rigid habitual behaviours. Within the Bayesian modelling approach of brain and behaviour, models have been proposed to perform planning as probabilistic inference but this approach encounters a crucial problem: explaining how such inference might be implemented in brain spiking networks. Recently, the literature has proposed some models that face this problem through recurrent spiking neural networks able to internally simulate state trajectories, the core function at the basis of planning. However, the proposed models have relevant limitations that make them biologically implausible, namely their world model is trained ‘off-line’ before solving the target tasks, and they are trained with supervised learning procedures that are biologically and ecologically not plausible. Here we propose two novel hypotheses on how brain might overcome these problems, and operationalise them in a novel architecture pivoting on a spiking recurrent neural network. The first hypothesis allows the architecture to learn the world model in parallel with its use for planning: to this purpose, a new arbitration mechanism decides when to explore, for learning the world model, or when to exploit it, for planning, based on the entropy of the world model itself. The second hypothesis allows the architecture to use an unsupervised learning process to learn the world model by observing the effects of actions. The architecture is validated by reproducing and accounting for the learning profiles and reaction times of human participants learning to solve a visuomotor learning task that is new for them. Overall, the architecture represents the first instance of a model bridging probabilistic planning and spiking-processes that has a degree of autonomy analogous to the one of real organisms., Author summary Goal-directed behaviour relies on brain processes supporting planning of actions based on their expected consequences before performing them in the environment. An important computational modelling approach proposes that the brain performs goal-directed processes on the basis of probability distributions and computations on them. A key challenge of this approach is to explain how these probabilistic processes can rely on the spiking processes of the brain. The literature has recently proposed some models that do so by ‘thinking ahead’ alternative possible action-outcomes based on low-level neuronal stochastic events. However, these models have a limited autonomy as they require to learn how the environment works (‘world model’) before solving the tasks, and use a biologically implausible learning process requiring an ‘external teacher’ to tell how their internal units should respond. Here we present a novel architecture proposing how organisms might overcome these challenging problems. First, the architecture can decide if exploring, to learn the world model, or planning, using such model, by evaluating how confident it is on the model knowledge. Second, the architecture can autonomously learn the world model based on experience. The architecture represents a first fully autonomous planning model relying on a spiking neural network.

An open question in neuroimaging is how to develop anatomical brain atlases for the analysis of functional data. Here, we present a cortical parcellation model based on macroanatomical information and test its validity on visuomotor-related cortical functional networks. The parcellation model is based on a recently developed cortical parameterization method (Auzias et al., [2013]: IEEE Trans Med Imaging 32:873-887), called HIP-HOP. This method exploits a set of primary and secondary sulci to create an orthogonal coordinate system on the cortical surface. A natural parcellation scheme arises from the axes of the HIP-HOP model running along the fundus of selected sulci. The resulting parcellation scheme, called MarsAtlas, complies with dorsoventral/rostrocaudal direction fields and allows inter-subject matching. To test it for functional mapping, we analyzed a MEG dataset collected from human participants performing an arbitrary visuomotor mapping task. Single-trial high-gamma activity, HGA (60-120 Hz), was estimated using spectral analysis and beamforming techniques at cortical areas arising from a Talairach atlas (i.e., Brodmann areas) and MarsAtlas. Using both atlases, we confirmed that visuomotor associations involve an increase in HGA over the sensorimotor and fronto-parietal network, in addition to medial prefrontal areas. However, MarsAtlas provided: (1) crucial functional information along both the dorsolateral and rostrocaudal direction; (2) an increase in statistical significance. To conclude, our results suggest that the MarsAtlas is a valid anatomical atlas for functional mapping, and represents a potential anatomical framework for integration of functional data arising from multiple techniques such as MEG, intracranial EEG and fMRI.

In social interactions, strategic uncertainty arises when the outcome of one’s choice depends on the choices of others. An important question is whether strategic uncertainty can be resolved by assessing subjective probabilities to the counterparts’ behavior, as if playing against nature, and thus transforming the strategic interaction into a risky (individual) situation. By means of functional magnetic resonance imaging with human participants we tested the hypothesis that choices under strategic uncertainty are supported by the neural circuits mediating choices under individual risk and deliberation in social settings (i.e. strategic thinking). Participants were confronted with risky lotteries and two types of coordination games requiring different degrees of strategic thinking of the kind ‘I think that you think that I think etc.’ We found that the brain network mediating risk during lotteries (anterior insula, dorsomedial prefrontal cortex and parietal cortex) is also engaged in the processing of strategic uncertainty in games. In social settings, activity in this network is modulated by the level of strategic thinking that is reflected in the activity of the dorsomedial and dorsolateral prefrontal cortex. These results suggest that strategic uncertainty is resolved by the interplay between the neural circuits mediating risk and higher order beliefs (i.e. beliefs about others’ beliefs). This work was supported by the European Research Council (ERC Consolidator Grant 617629 to G.C.); Human Frontier Science Program Organization (for life science and physics) to G.C., A.B. and R.N.; Dirección General de Investigación Cientifíca y Técnica (ECO2011-25295) (ECO2008-01768), Barcelona Graduate School of Economics and the Generalitat de Catalunya to R.N. A.B. was also supported by the Projet Exploratoire Premier Soutien (PEPS) Bio-Maths-Info (BMI) of the CNRS–INRIA–INSERM.

Cognitive functions arise from the coordination of large-scale brain networks. However, the principles governing interareal functional connectivity dynamics (FCD) remain elusive. Here, we tested the hypothesis that human executive functions arise from the dynamic interplay of multiple networks. To do so, we investigated FCD mediating a key executing function, known as arbitrary visuomotor mapping, using brain connectivity analyses of high-gamma activity recorded using MEG and intracranial EEG. Visuomotor mapping was found to arise from the dynamic interplay of three partly overlapping cortico-cortical and cortico-subcortical functional connectivity (FC) networks. First, visual and parietal regions coordinated with sensorimotor and premotor areas. Second, the dorsal frontoparietal circuit together with the sensorimotor and associative frontostriatal networks took the lead. Finally, cortico-cortical interhemispheric coordination among bilateral sensorimotor regions coupled with the left frontoparietal network and visual areas. We suggest that these networks reflect the processing of visual information, the emergence of visuomotor plans, and the processing of somatosensory reafference or action's outcomes, respectively. We thus demonstrated that visuomotor integration resides in the dynamic reconfiguration of multiple cortico-cortical and cortico-subcortical FC networks. More generally, we showed that visuomotor-related FC is nonstationary and displays switching dynamics and areal flexibility over timescales relevant for task performance. In addition, visuomotor-related FC is characterized by sparse connectivity with density SIGNIFICANCE STATEMENTExecutive functions are supported by the dynamic coordination of neural activity over large-scale networks. The properties of large-scale brain coordination processes, however, remain unclear. Using tools combining MEG and intracranial EEG with brain connectivity analyses, we provide evidence that visuomotor behaviors, a hallmark of executive functions, are mediated by the interplay of multiple and spatially overlapping subnetworks. These subnetworks span visuomotor-related areas, the cortico-cortical and cortico-subcortical interactions of which evolve rapidly and reconfigure over timescales relevant for behavior. Visuomotor-related functional connectivity dynamics are characterized by sparse connections, nonstationarity, switching dynamics, and areal flexibility. We suggest that these properties represent key aspects of large-scale functional networks and cognitive architectures.

International audience; The reference electrophysiological pattern at seizure onset is the "rapid discharge", as visible on intracerebral EEG. This discharge typically corresponds to a decrease of synchrony across brain areas. In contrast, the preictal period can exhibit patterns of increased synchrony, which can be quantified by network measures. Our objective was to compare preictal synchrony with a quantification of the rapid discharge as provided by the Epileptogenicity Index (EI).We investigated 24 seizures from 12 patients recorded by stereotaxic EEG (SEEG). Seizures were classified visually as containing preictal synchrony or not. We computed pairwise nonlinear correlation (h2) across channels in the 8s preceding the rapid discharge. The sum of ingoing and outgoing links (IN and OUT node strength), as well as the sum of all links (total strength, TOT) were computed for each region. We tested several filtering schemes, and quantified the capacity of each strength measure to serve as a detector of regions with high EI values using a receiver operating characteristic (ROC) analysis.We found that the best correspondence between node strength and EI was obtained for the OUT and TOT measures, for signals filtered in the 15-40 Hz band - i.e. for the band corresponding to the spiky part of epileptic discharges. In agreement with these results, we also found that the ROC results were improved when considering only seizures with visible synchronous patterns in the preictal period. Our results suggest that measuring strength of preictal connectivity graphs can bring useful clinical information on the epileptogenic zone, and potentially give more relevant information than the number of interictal spikes.

Associative theory postulates that learning the consequences of our actions in a given context is represented in the brain as stimulus-response-outcome associations that evolve according to prediction-error signals (the discrepancy between the observed and predicted outcome). We tested the theory on brain functional magnetic resonance imaging data acquired from human participants learning arbitrary visuomotor associations. We developed a novel task that systematically manipulated learning and induced highly reproducible performances. This granted the validation of the model-based results and an in-depth analysis of the brain signals in representative single trials. Consistent with the Rescorla-Wagner model, prediction-error signals are computed in the human brain and selectively engage the ventral striatum. In addition, we found evidence of computations not formally predicted by the Rescorla-Wagner model. The dorsal fronto-parietal network, the dorsal striatum, and the ventrolateral prefrontal cortex are activated both on the incorrect and first correct trials and may reflect the processing of relevant visuomotor mappings during the early phases of learning. The left dorsolateral prefrontal cortex is selectively activated on the first correct outcome. The results provide quantitative evidence of the neural computations mediating arbitrary visuomotor learning and suggest new directions for future computational models.

An open question in neuroimaging is how to develop anatomical brain atlases for the analysis of functional data. Here, we present a cortical parcellation model based on macroanatomical information and test its validity on visuomotor‐related cortical functional networks. The parcellation model is based on a recently developed cortical parameterization method (Auzias et al., [2013]: IEEE Trans Med Imaging 32:873–887), called HIP‐HOP. This method exploits a set of primary and secondary sulci to create an orthogonal coordinate system on the cortical surface. A natural parcellation scheme arises from the axes of the HIP‐HOP model running along the fundus of selected sulci. The resulting parcellation scheme, called MarsAtlas, complies with dorsoventral/rostrocaudal direction fields and allows inter‐subject matching. To test it for functional mapping, we analyzed a MEG dataset collected from human participants performing an arbitrary visuomotor mapping task. Single‐trial high‐gamma activity, HGA (60–120 Hz), was estimated using spectral analysis and beamforming techniques at cortical areas arising from a Talairach atlas (i.e., Brodmann areas) and MarsAtlas. Using both atlases, we confirmed that visuomotor associations involve an increase in HGA over the sensorimotor and fronto‐parietal network, in addition to medial prefrontal areas. However, MarsAtlas provided: (1) crucial functional information along both the dorsolateral and rostrocaudal direction; (2) an increase in statistical significance. To conclude, our results suggest that the MarsAtlas is a valid anatomical atlas for functional mapping, and represents a potential anatomical framework for integration of functional data arising from multiple techniques such as MEG, intracranial EEG and fMRI. Hum Brain Mapp 37:1573‐1592, 2016. © 2016 Wiley Periodicals, Inc.

Current learning theory provides a comprehensive description of how humans and other animals learn, and places behavioral flexibility and automaticity at heart of adaptive behaviors. However, the computations supporting the interactions between goal-directed and habitual decision-making systems are still poorly understood. Previous functional magnetic resonance imaging (fMRI) results suggest that the brain hosts complementary computations that may differentially support goal-directed and habitual processes in the form of a dynamical interplay rather than a serial recruitment of strategies. To better elucidate the computations underlying flexible behavior, we develop a dual-system computational model that can predict both performance (i.e., participants' choices) and modulations in reaction times during learning of a stimulus–response association task. The habitual system is modeled with a simple Q-Learning algorithm (QL). For the goal-directed system, we propose a new Bayesian Working Memory (BWM) model that searches for information in the history of previous trials in order to minimize Shannon entropy. We propose a model for QL and BWM coordination such that the expensive memory manipulation is under control of, among others, the level of convergence of the habitual learning. We test the ability of QL or BWM alone to explain human behavior, and compare them with the performance of model combinations, to highlight the need for such combinations to explain behavior. Two of the tested combination models are derived from the literature, and the latter being our new proposal. In conclusion, all subjects were better explained by model combinations, and the majority of them are explained by our new coordination proposal.

While neuroscience research has tremendously advanced our knowledge about the neural mechanisms of individual learning, i.e. through trial-and-error, it is only recently that neuroscientists have begun to study observational learning, and thus little is known about its neural mechanisms. One limitation is that observational learning has been addressed under unconstrained experimental conditions, not compatible with neuronal recordings. This study examined observational learning in macaque monkeys under the constraining conditions of behavioral neurophysiology. Two animals sat in primate chairs facing each other, with their head fixed. A touch screen was placed face up between the chairs at arm's reach, and the monkeys were trained on an abstract visuomotor associative task. In one experiment, the monkeys alternated the roles of "actor" and "observer". The actor learned to associate visual cues with reaching targets, while the observer "watched" freely. Then, the observer was given the same cue-target associations just performed by the actor, or had to learn new, not previously observed ones. The results show that learning performance is better after observation. In experiment 2, one monkey learned from a human actor who performed the task with errors only, or with successes only in separate blocks. The monkey's gain in performance was higher after observation of errors than after successes. The findings suggest that observational learning can occur even under highly constraining conditions, and open the way for investigating the neuronal correlates of social learning using the methods of behavioral neurophysiology.

Adaptive behaviors are built on the arbitrary linkage of sensory inputs to actions and goals. Although the sensorimotor and associative frontostriatal circuits are known to mediate arbitrary visuomotor mappings, the underlying corticocortico dynamics remain elusive. Here, we take a novel approach exploiting gamma-band neural activity to study the human cortical networks and corticocortical functional connectivity mediating arbitrary visuomotor mapping. Single-trial gamma-power time courses were estimated for all Brodmann areas by combing magnetoencephalographic and MRI data with spectral analysis and beam-forming techniques. Linear correlation and Granger causality analyses were performed to investigate functional connectivity between cortical regions. The performance of visuomotor associations was characterized by an increase in gamma-power and functional connectivity over the sensorimotor and frontoparietal network, in addition to medial prefrontal areas. The superior parietal area played a driving role in the network, exerting Granger causality on the dorsal premotor area. Premotor areas acted as relay from parietal to medial prefrontal cortices, which played a receiving role in the network. Link community analysis further revealed that visuomotor mappings reflect the coordination of multiple subnetworks with strong overlap over motor and frontoparietal areas. We put forward an associative account of the underlying cognitive processes and corticocortical functional connectivity. Overall, our approach and results provide novel perspectives toward a better understanding of how distributed brain activity coordinates adaptive behaviors.SIGNIFICANCE STATEMENTIn everyday life, most of our behaviors are based on the arbitrary linkage of sensory information to actions and goals, such as stopping at a red traffic light. Despite their automaticity, such behaviors rely on the activity of a large brain network and elusive interareal functional connectivity. We take a novel approach exploiting noninvasive recordings of human brain activity to reveal the cortical networks and corticocortical functional connectivity mediating visuomotor mappings. Parietal areas were found to play a driving role in the network, whereas premotor areas acted as relays from parietal to medial prefrontal cortices, which played a receiving role. Overall, our approach and results provide novel perspectives toward a better understanding of how distributed brain activity coordinates adaptive behaviors.

In the present work a simple technique for fMRI data analysis is presented. Artifacts due to random and stimulus-correlated motions are corrected without image registration procedures. The first step of our procedure is the calculation of the raw activation map by correlation analysis. The task related motion artifacts arise at the tissue interfaces, including vessels: when image intensity gradient is calculated the high values correspond to interface regions. To eliminate stimulus-correlated motion artifacts the intensity gradient image, obtained from the fMRI data set, is compared to the raw activation map. Since small random motions decrease the value of the correlation coefficient (R) of the external pixels of the activation areas, in the last step of our analysis procedures the clusters are extended to connected pixels having R values smaller than the defined threshold. Each cluster is expanded until the R value of the cluster average intensity is kept constant. The procedure has been tested with both GRE and EPI studies. The presented approach is a fast and robust technique useful for preliminary or on-line analysis of fMRI data. © 2002 Elsevier Science Inc. All rights reserved.

Contemporary behavioral learning theory provides a comprehensive description of how we and other animals learn, and places behavioral flexibility and automaticity at heart of adaptive behaviors. However, to our knowledge, the computations supporting the interactions between deliberative and habitual decision-making systems are still poorly understood. Previous functional magnetic resonance imaging (fMRI) results suggest that the dorsal striatum host complementary computations that may differentially support deliberative and habitual processes [1] in the form of a dynamical interplay rather than a serial recruitment of strategies. From the same instrumental task, we develop a dual-system computational model of the two systems that can predict both performance (i.e., participant choices) and modulations in reaction times during learning. The instrumental task is a trial-and-error learning task requiring participants to find the correct associations between color stimuli and finger responses. To model the habitual system, we use a simple Q-learning algorithm (QL) [2] whose properties are fast responses, but slow convergence. For the deliberative (i.e goal-directed) system, we propose a new Bayesian Working Memory (BWM) which searches for information in the history of previous trials and stops as soon as the uncertainty on the action to perform decreases below a certain threshold. Last, we also propose a model for QL and BWM coordination. Currently, most models of system selection tend to control action selection concurrently, using either the deliberative or habitual model according to uncertainty criteria [3,4]. Only one model has investigated the relation between working memory and reinforcement learning [5] without, however explicitly modeling the temporal aspect of memory manipulation. In our approach, we propose a model for QL and BWM coordination. QL and BWM are merged such that the expensive memory manipulation is under control of, among others, the level of convergence of the habitual learning. Consequently, we also predict specific reaction times for each model that can be compared with the evolution of reaction times in instrumental learning tasks. Models are optimized for each subject with the NSGA-2 multi-objective evolutionary algorithm. The first fitness function is the Bayesian Information Criterion for individual choices. The second fitness function is also a likelihood that maximizes the probability of performing reaction times similar to humans. We compare the ability of the new model to explain human behavior with the QL or BWM only, as well as with a combination of these models based on [4], which reveals that the proposed model is in general more accurate. To conclude, we suggest that a close combination of BWM and QL better explains both choices and reaction times for most participants.

International audience; We have previously shown that mental rehearsal can replace up to 75% of physical practice for learning a visuomotor task (Allami, Paulignan, Brovelli, & Boussaoud, (2008). Experimental Brain Research,184,105-113). Presumably, mental rehearsal must induce brain changes that facilitate motor learning. We tested this hypothesis by recording scalp electroencephalographic activity (EEG) in two groups of subjects. In one group, subjects executed a reach to grasp task for 240 trials. In the second group, subjects learned the task through a combination of mental rehearsal for the initial 180 trials followed by the execution of 60 trials. Thus, one group physically executed the task for 240 trials, the other only for 60 trials. Amplitudes and latencies of event-related potentials (ERPs) were compared across groups at different stages during learning. We found that ERP activity increases dramatically with training and reaches the same amplitude over the premotor regions in the two groups, despite large differences in physically executed trials. These findings suggest that during. mental rehearsal, neuronal changes occur in the motor networks that make physical practice after mental rehearsal more effective in configuring functional networks for skilful behaviour. (C) 2013 Elsevier Ltd. All rights reserved.

EEG/VEP and fMRI responses to periodic visual stimulation are reported. The purpose of these experiments was to look for similar patterns in the time series produced by each method to help understand the relationship between the two. The stimulation protocol was the same for both sets of experiments and consisted of five complete cycles of checkerboard pattern reversal at 1.87 Hz for 30 s followed by 30 s of a stationary checkerboard. The fMRI data was analyzed using standard methods, while the EEG was analyzed with a new measurement of activation-the VEPEG. Both VEPEG and fMRI time series contain the fundamental frequency of the stimulus and quasi harmonic components-an unexplained double frequency commonly found in fMRI data. These results have prompted a reappraisal of the methods for analyzing fMRI data and have suggested a connection between our findings and much older published invasive electrophysiological measurements of blood flow and the partial pressures of oxygen and carbon dioxide. Overall our new analysis suggests that fMRI signals are strongly dependant on hydraulic blood flow effects. We distinguish three categories of fMRI signal corresponding to: focal activated regions of brain tissue; diffuse nonspecific regions of steal; and major cerebral vessels of arterial supply or venous drainage. Each category of signal has its own finger print in frequency, amplitude, and phase. Finally, we put forward the hypothesis that modulations in blood flow are not only the consequence but are also the cause of modulations in functional activity.

International audience; Since the first descriptions of sensorimotor rhythms by Berger (1929) and by Jasper and Penfield (1949), the potential role of beta oscillations (~13–30 Hz) in the brain has been intensely investigated. We start this review by showing that experimental studies in humans and monkeys have reached a consensus on the facts that sensorimotor beta power is low during movement, transiently increases after movement end (the " beta rebound ") and tonically increases during object grasping. Recently, a new surge of studies exploiting more complex sensorimotor tasks including multiple events, such as instructed delay tasks, reveal novel characteristics of beta oscillatory activity. We therefore proceed by critically reviewing also this literature to understand whether modulations of beta oscillations in task epochs other than those during and after movement are consistent across studies, and whether they can be reconciled with a role for beta oscillations in sensorimotor transmission. We indeed find that there are additional processes that also strongly affect sen-sorimotor beta oscillations, such as visual cue anticipation and processing, fitting with the view that beta oscillations reflect heightened sensorimotor transmission beyond somatosensation. However, there are differences among studies, which may be interpreted more readily if we assume multiple processes, whose effects on the overall measured beta power overlap in time. We conclude that beta oscillations observed in sensorimotor cortex may serve large-scale communication between sensorimotor and other areas and the periphery.

Learning what behaviour is appropriate in a specific context by observing the actions of others and their outcomes is a key constituent of human cognition, because it saves time and energy and reduces exposure to potentially dangerous situations. Observational learning of associative rules relies on the ability to map the actions of others onto our own, process outcomes, and combine these sources of information. Here, we combined newly developed experimental tasks and functional magnetic resonance imaging (fMRI) to investigate the neural mechanisms that govern such observational learning. Results show that the neural systems involved in individual trial-and-error learning and in action observation and execution both participate in observational learning. In addition, we identified brain areas that specifically activate for others' incorrect outcomes during learning in the posterior medial frontal cortex (pMFC), the anterior insula and the posterior superior temporal sulcus (pSTS).

International audience; Granger causality analysis is becoming central for the analysis of interactions between neural populations and oscillatory networks. However, it is currently unclear whether single-trial estimates of Granger causality spectra can be used reliably to assess directional influence. We addressed this issue by combining single-trial Granger causality spectra with statistical inference based on general linear models. The approach was assessed on synthetic and neurophysiological data. Synthetic bivariate data was generated using two autoregressive processes with unidirectional coupling. We simulated two hypothetical experimental conditions: the first mimicked a constant and unidirectional coupling, whereas the second modelled a linear increase in coupling across trials. The statistical analysis of single-trial Granger causality spectra, based on t-tests and linear regression, successfully recovered the underlying pattern of directional influence. In addition, we characterised the minimum number of trials and coupling strengths required for significant detection of directionality. Finally, we demonstrated the relevance for neurophysiology by analysing two local field potentials (LFPs) simultaneously recorded from the prefrontal and premotor cortices of a macaque monkey performing a conditional visuomotor task. Our results suggest that the combination of single-trial Granger causality spectra and statistical inference provides a valuable tool for the analysis of large-scale cortical networks and brain connectivity.

Functional magnetic resonance imaging (fMRI) exploits blood-oxygen-level-dependent (BOLD) contrasts to map neural activity associated with a variety of brain functions including sensory processing, motor control, and cognitive and emotional functions. The general linear model (GLM) approach is used to reveal task-related brain areas by searching for linear correlations between the fMRI time course and a reference model. One of the limitations of the GLM approach is the assumption that the covariance across neighbouring voxels is not informative about the cognitive function under examination. Multivoxel pattern analysis (MVPA) represents a promising technique that is currently exploited to investigate the information contained in distributed patterns of neural activity to infer the functional role of brain areas and networks. MVPA is considered as a supervised classification problem where a classifier attempts to capture the relationships between spatial pattern of fMRI activity and experimental conditions. In this paper , we review MVPA and describe the mathematical basis of the classification algorithms used for decoding fMRI signals, such as support vector machines (SVMs). In addition, we describe the workflow of processing steps required for MVPA such as feature selection, dimensionality reduction, cross-validation, and classifier performance estimation based on receiver operating characteristic (ROC) curves.

Observational learning allows individuals to acquire knowledge without incurring in the costs and risks of discovering and testing. The neural mechanisms mediating the retrieval of rules learned by observation are currently unknown. To explore this fundamental cognitive ability, we compared the brain responses when retrieving visuomotor associations learned either by observation or by individual learning. To do so, we asked eleven adults to learn two sets of arbitrary visuomotor associations: one set was learned through the observation of an expert actor while the other was learned by trial and error. During fMRI scanning, subjects were requested to retrieve the visuomotor associations previously learned under the two modalities. The conjunction analysis between the two learning conditions revealed a common brain network that included the ventral and dorsal lateral prefrontal cortices, the superior parietal lobe and the pre-SMA. This suggests the existence of a mirror-like system responsible for the storage of rules learned either by trial and error or by observation of others' actions. In addition, the pars triangularis in the right prefrontal cortex (BA45) was found to be selective for rules learned by observation. This suggests a preferential role of this area in the storage of rules learned in a social context.

International audience; Sports psychology suggests that mental rehearsal facilitates physical practice in athletes and clinical rehabilitation attempts to use mental rehearsal to restore motor function in hemiplegic patients. Our aim was to examine whether mental rehearsal is equivalent to physical learning, and to determine the optimal proportions of real execution and rehearsal. Subjects were asked to grasp an object and insert it into an adapted slot. One group (G0) practiced the task only by physical execution (240 trials); three groups imagined performing the task in different rates of trials (25%, G25; 50%, G50; 75%, G75), and physically executed movements for the remaining trials; a fourth, control group imagined a visual rotation task in 75% of the trials and then performed the same motor task as the others groups. Movement time (MT) was compared for the first and last physical trials, together with other key trials, across groups. All groups learned, suggesting that mental rehearsal is equivalent to physical motor learning. More importantly, when subjects rehearsed the task for large numbers of trials (G50 and G75), the MT of the first executed trial was significantly shorter than the first executed trial in the physical group (G0), indicating that mental practice is better than no practice at all. Comparison of the first executed trial in G25, G50 and G75 with the corresponding trials in G0 (61, 121 and 181 trials), showed equivalence between mental and physical practice. At the end of training, the performance was much better with high rates of mental practice (G50/G75) compared to physical practice alone (G0), especially when the task was difficult. These findings confirm that mental rehearsal can be beneficial for motor learning and suggest that imagery might be used to supplement or partly replace physical practice in clinical rehabilitation.

Successful adaptation relies on the ability to learn the consequence of our actions in different environments. However, understanding the neural bases of this ability still represents one of the great challenges of system neuroscience. In fact, the neuronal plasticity changes occurring during learning cannot be fully controlled experimentally and their evolution is hidden. Our approach is to provide hypotheses about the structure and dynamics of the hidden plasticity changes using behavioral learning theory. In fact, behavioral models of animal learning provide testable predictions about the hidden learning representations by formalizing their relation with the observables of the experiment (stimuli, actions and outcomes). Thus, we can understand whether and how the predicted learning processes are represented at the neural level by estimating their evolution and correlating them with neural data. Here, we present a bayesian model approach to estimate the evolution of the internal learning representations from the observations of the experiment (state estimation), and to identify the set of models' parameters (parameter estimation) and the class of behavioral model (model selection) that are most likely to have generated a given sequence of actions and outcomes. More precisely, we use Sequential Monte Carlo methods for state estimation and the maximum likelihood principle (MLP) for model selection and parameter estimation. We show that the method recovers simulated trajectories of learning sessions on a single-trial basis and provides predictions about the activity of different categories of neurons that should participate in the learning process. By correlating the estimated evolutions of the learning variables, we will be able to test the validity of different models of instrumental learning and possibly identify the neural bases of learning.

Visuomotor transformation processes are essential when accurate reaching movements towards a visual target have to be performed. In contrast, those transformations are not needed for similar, but non-visually guided, arm movements. According to previous studies, these transformations are carried out by neuronal populations located in the parietal and frontal cortical areas (the so-called “dorsal visual stream”). However, it is still debated whether these processes are mediated by the sequential and/or parallel activation of the frontoparietal areas. To investigate this issue, we designed a task where the same visual cue could represent either the target of a reaching/pointing movement or the go-signal for a similar but non-targeting arm movement. By subtracting the event-related potentials (ERPs) recorded from healthy subjects performing the two conditions, we identified the brain processes underlying the visuomotor transformations needed for accurate reaching/pointing movements. We then localized the generators by means of cortical current density (CCD) reconstruction and studied their dynamics from visual cue presentation to movement onset. The results showed simultaneous activation of the parietal and frontal areas from 140 to 260 ms. The results are interpreted as neural correlates of two critical phases of visuomotor integration, namely target selection and movement selection. Our findings suggest that the visuomotor transformation processes required for correct reaching/pointing movements do not rely on a purely sequential activation of the frontoparietal areas, but mainly on a parallel information processing system, where feedback circuits play an important role before movement onset.

International audience; The premotor cortex is well known for its role in motor planning. In addition, recent studies have shown that it is also involved in nonmotor functions such as attention and memory, a notion derived from both animal neurophysiology and human functional imaging. The present study is an attempt to bridge the gap between these experimental techniques in the human brain, using a task initially designed to dissociate attention from intention in the monkey, and recently adapted for a functional magnetic resonance imaging (fMRI) study [Simon, S.R., Meunier, M., Piettre, L., Berardi, A.M., Segebarth, C.M., Boussaoud, D. (2002). Spatial attention and memory versus motor preparation: premotor cortex involvement as revealed by fMRI. J. Neurophysiol., 88, 2047-57]. Intracranial EEG was recorded from the cortical regions preferentially active in the spatial attention and/or working memory task and those involved in motor intention. The results show that, among the different intracranial EEG responses, only the high gamma frequency (60-200 Hz) oscillatory activity both dissociates attention/memory from motor intention and spatially colocalizes with the fMRI-identified premotor substrates of these two functions. This finding provides electrophysiological confirmation that the human premotor cortex is involved in spatial attention and/or working memory. Additionally, it provides timely support to the idea that high gamma frequency oscillations are involved in the cascade of neural processes underlying the hemodynamic responses measured with fMRI [Logothetis, N.K., Pauls, J., Augath, M., Trinath, T. and Oeltermann, A. (2001). Neurophysiological investigation of the basis of the fMRI signal. Nature, 412, 150-7], and suggests a functional selectivity of the gamma oscillations that could be critical for future EEG investigations, whether experimental or clinical.

Previous studies have shown that synchronized beta frequency (14-30 Hz) oscillations in the primary motor cortex are involved in maintaining steady contractions of contralateral arm and hand muscles. However, little is known about the role of postcentral cortical areas in motor maintenance and their patterns of interaction with motor cortex. We investigated the functional relations of beta-synchronized neuronal assemblies in pre- and postcentral areas of two monkeys as they pressed a hand lever during the wait period of a visual discrimination task. By using power and coherence spectral analysis, we identified a beta-synchronized large-scale network linking pre- and postcentral areas. We then used Granger causality spectra to measure directional influences among recording sites. In both monkeys, strong Granger causal influences were observed from primary somatosensory cortex to both motor cortex and inferior posterior parietal cortex, with the latter area also exerting Granger causal influences on motor cortex. Granger causal influences from motor cortex to postcentral sites, however, were weak in one monkey and not observed in the other. These results are the first, to our knowledge, to demonstrate in awake monkeys that synchronized beta oscillations bind multiple sensorimotor areas into a large-scale network during motor maintenance behavior and carry Granger causal influences from primary somatosensory and inferior posterior parietal cortices to motor cortex.

In order to assess the role played by area V6A in visuomotor control, two adult green monkeys ( Cercopithecus aethiops) were subjected to small, bilateral lesions in the anterior bank of the parieto-occipital sulcus. Before and after the lesions, monkeys were tested for naturally designed reaching, grasping and picking-up pieces of food from various positions on a plate and from a differently oriented narrow slit. All movements were recorded with closed circuit TV and analysed offline on a single-photogram basis for defective reaching and wrist orientation. V6A lesions provoked parietal weakness, reluctance to move, and specific deficits in reaching, wrist orientation and grasping. Recovery from the observed deficits was rapid, even after a second, contralateral lesion was given, creating a bilateral lesion. Thus, together with previous anatomical and electrophysiological data, these results directly support the hypothesis that area V6A is part of the network involved in the control of reaching movements and wrist orientation.

International audience; The relationship between network structure and dynamics is one of the most extensively investigated problems in the theory of complex systems of recent years. Understanding this relationship is of relevance to a range of disciplines—from neuroscience to geomorphology. A major strategy of investigating this relationship is the quantitative comparison of a representation of network architecture (structural connectivity, SC) with a (network) representation of the dynamics (functional connectivity, FC). Here, we show that one can distinguish two classes of functional connectivity—one based on simultaneous activity (co-activity) of nodes, the other based on sequential activity of nodes. We delineate these two classes in different categories of dynamical processes—excitations, regular and chaotic oscillators—and provide examples for SC/FC correlations of both classes in each of these models. We expand the theoretical view of the SC/FC relationships, with conceptual instances of the SC and the two classes of FC for various application scenarios in geomorphology, ecology, systems biology, neuroscience and socio-ecological systems. Seeing the organisation of dynamical processes in a network either as governed by co-activity or by sequential activity allows us to bring some order in the myriad of observations relating structure and function of complex networks.

In the field of instrumental learning, mammals are able to implement two different behavioral strategies to interact with the environment: goal directed behavior (GDB), computationally flexible but slow, suitable to learn new tasks and adapt to changing environments; and habitual behavior, hard-coded, but suitable for faster motor responses and facing recurrent tasks. The advantage of GDB resides in the use of an inner representation of the environment, a ‘model of the world’, to encode stimuli-actions-outcomes associations, and its exploitation to choose future actions, in a process called planning. GDB is supported by large-scale networks involving both cortical and subcortical regions. Nevertheless, several open questions still remain. The aim of this thesis is to contribute to the understanding of three open questions (declined in three studies) that pertain to the neural and computational mechanisms of GDB.In the first study, we investigated how complex computations, such as learning the model of the world and planning, can emerge from simple neural activity. To achieve that, we built a spiking neural network, able to encode stimulus-actions-outcomes associations as a hidden Markov model (HMM), using biologically inspired mechanisms such as spike-timing dependent plasticity (STDP), and to test this model to correctly plan actions in order to solve a visuomotor goal directed task. The performance of the model was validated on behavioral data from human participants that performed the same task.In the second study, we assessed the importance of striatum in encoding the reward prediction error (RPE) signals, a relevant update signal in most instrumental learning models. To do so, we analysed local field potentials (LFPs) recorded in rhesus macaque striatum while performing a probabilistic goal-directed learning task. Then, we computed the trial-by-trial RPE using a Q-learning model fitted on monkeys’ behavior. Our results showed a significant increase of mutual information (MI) between the beta-band (15-30Hz) oscillatory activity and the RPE after the outcome presentation. Moreover, such correlates of RPE signals form an anatomo-functional gradient in the striatum, showing stronger effects toward the rostro-ventral part and vanishing toward the caudo-dorsal part.In the third study, we investigated the neural correlates of GDB at the whole-brain cortical level in humans. To do so, we recorded the brain activity of human participants using magnetoencephalography (MEG) while they were performing a goal-directed causal learning task. We exploited cortical high-gamma activity (HGA, 60-120Hz) to map the spatio-temporal dynamics during learning. In particular, we used an ideal observer Bayesian model to estimate the trial-by-trial evolution of relevant behavioral variables, such as action-outcome probabilities and contingency values. We used MI and group-level cluster-based statics between HGA and those variables to obtain a whole brain profile of behavioral-dependent regions of interests’ activity, confirming some results from the literature., Dans le domaine de l'apprentissage instrumental, les mammifères sont capables de mettre en œuvre deux stratégies comportementales différentes pour interagir avec l'environnement: le comportement dirigé vers un but (“goal-directed behavior”, GDB), flexible sur le plan computationnel mais lent, adapté à l'apprentissage de nouvelles tâches et à l'adaptation à des environnements changeants; et le comportement habituel, encodé de façon rigide, mais adapté à des réponses motrices plus rapide, adapté aux tâches récurrentes. L'avantage du GDB réside dans l'utilisation d'une représentation interne de l'environnement, un ‘modèle du monde’, pour encoder les associations stimuli-actions-conséquences, et dans l'utilisation de ce modèle pour choisir les actions futures au cours du processus de planification. Le GDB est soutenu par des réseaux cérébraux à grande échelle impliquant des régions corticales et sous-corticales. Néanmoins, plusieurs questions ouvertes demeurent. L'objectif de cette thèse est de contribuer à la compréhension de trois questions ouvertes (déclinées en trois études) qui concernent les mécanismes neuronaux et computationnels du GDB.Dans une première étude, nous avons cherché à savoir comment des calculs complexes, tels que l'apprentissage du modèle du monde et la planification, peuvent émerger de l’activité neuronale. Pour ce faire, nous avons construit un réseau de neurones actifs, capable d'encoder des associations stimulus-actions-conséquences sous la forme d'un modèle de Markov caché (Hidden Markov Model, HMM), en utilisant des mécanismes d'inspiration biologique tels que la ‘spike-timing dependent plasticity’ (STDP), et d'utiliser ce modèle pour planifier correctement des actions afin de résoudre une tâche visuomotrice. Les performances du modèle ont été validées sur des données comportementales de participants humains ayant effectué la même tâche.Dans une deuxième étude, nous avons évalué l'importance du striatum dans l'encodage de l'erreur de prédiction de la récompense (Reward Prediction Error, RPE), un signal de mise à jour pertinent dans la plupart des modèles d'apprentissage instrumental. Pour ce faire, nous avons analysé les potentiels de champ locaux (Local Field Potentials, LFP) enregistrés dans le striatum de macaques rhésus pendant l'exécution d'une tâche d'apprentissage probabiliste dirigée vers un but. Ensuite, nous avons calculé la RPE essai par essai en utilisant un modèle de ‘Q-learning’ adapté au comportement des singes. Nos résultats ont montré une augmentation significative de l'information mutuelle (Mutual Information, MI) entre l'activité oscillatoire dans la bande bêta (15-30 Hz) et la RPE après le résultat de l’action. De plus l'information sur la RPE forme un gradient impliquant l'ensemble du striatum, plus intense dans la partie rostro-ventrale que dans la partie caudo-dorsale.Dans la troisième étude, nous avons étudié les corrélats neuronaux du GDB au niveau cortical du cerveau entier chez l'homme. Pour ce faire, nous avons enregistré l'activité corticale de participants humains à l'aide de la magnétoencéphalographie (MEG) pendant qu'ils effectuaient une tâche d'apprentissage causal dirigée vers un but. Nous nous sommes concentrés sur l'extraction et l'analyse de l'activité oscillatoire dans la bande gamma haute (High-Gamma Activity, HGA 60-120 Hz) pour mapper la dynamique spatio-temporelle pendant l'apprentissage. Ensuite, nous avons utilisé un modèle Bayésien d'observateur idéal pour estimer l'évolution essai par essai des variables comportementales pertinentes, telles que les probabilités de résultats d'action et les valeurs de contingence. Nous avons utilisé la MI et des statiques au niveau du groupe basées sur le cluster entre le HGA et ces variables pour obtenir un profil du cerveau entier de l'activité des régions d'intérêt dépendant du comportement, confirmant certains résultats de la littérature.