Jean-Marc Fellous - US grants

Considerable research demonstrates that people have a strong tendency to use facial appearance when forming first impressions of others' psychological traits, and that these impressions show considerable consensus across perceivers. The aim of the proposed research is to explain consensual first impressions of faces. The explanation to be tested is that social attributes revealed by the facial qualities that mark babies, fitness, or emotion are overgeneralized to people whose facial structure resembles that of babies, a particular level of fitness, or a particular emotional state. Research investigating relationships between facial qualities and trait impressions has provided indirect support for the overgeneralization hypothesis. However, the assumption that these impressions actually happen because of the overgeneralization of reactions to faces has not been tested. Connectionist modeling will be used to test the overgeneralization hypothesis. This technique may reveal whether the physical similarity between two faces is sufficient to account for similar impressions of them independent of similarities in the social or semantic implications of the faces. A series of experiments will provide facial metrics as input to standard back-propagation neural networks. Additional experiments will be conducted to determine whether human perceivers' impressions of the faces can be predicted from the responses of the networks. The proposed experiments will test the overgeneralization explanations for consensual first impressions of faces, and they will help identify the qualities that distinguish among faces that vary in emotional expression, genetic fitness, and maturity. The research will contribute to a better understanding of how people form impressions of one another, and ultimately how such impressions influence social interaction and interpersonal relationships.

Understanding how memory is encoded and maintained in our brain is paramount to understanding cognitive functions. Unlike in a computer, human memories are continuously consolidated, reconsolidated, and integrated within the context of what has already been learned. This process is thought to involve exchanges of information between the cortex and the hippocampus during sleep. The investigators will study the ability of small groups of cells in the rodent hippocampus and medial prefrontal cortex (mPFC) to become transiently co-active during sleep periods occurring many hours after learning has taken place. Rats will be engaged in learning tasks aimed at selectively activating one or both of these areas. It is expected that the activity recorded during post-task sleep will be correlated with the activity of the same cells during the task in a manner compatible with the nature of the task and the specifics of the learning. This type of reactivation is considered to be a basic mechanism for memory consolidation.

The investigators have developed new analytical tools based on fuzzy clustering and information geometry. Preliminary data show that short episodes of reactivation occur with different time courses in these two structures, as is often proposed on theoretical grounds. In this project, the investigators will study how this reactivation is coordinated across two connected brain areas (CA3-CA1, CA1-mPFC) on very long time scales, and how single neurons contribute to single reactivating episodes.

These studies will yield insights into the long-term temporal and spatial dynamics of reactivation in the adult rodent. They will also contribute to a better understanding of the neural basis of memory consolidation and reconsolidation in cortex and hippocampus, and the relationship between memory consolidation and sleep.

Spatial navigation is a complex cognitive process that relies on robust and adaptive mechanisms to relate current and future spatial positions to specific locations in the environment. The goal of this project is to provide a better understanding of spatial navigation by integrating information obtained from experimental studies in rats, computational models, and experiments on robots that will test new hypotheses on how these mechanisms work.

The hippocampus and medial entorhinal cortex (MEC) are major brain regions involved in mammalian spatial navigation. While the role of place cells in the hippocampus has been extensively studied, there are still many open questions on the functional role of MEC grid cells and their interaction with the hippocampal place cells. Of interest to this proposal is the recent finding that grid cells are organized in an orderly fashion along the dorso-ventral axis of the MEC, with dorsal grids being much more tightly spaced than ventral ones. The investigators hypothesize that this multiscale organization endows the navigation system with a coding mechanism that will inherently achieve robustness with respect to external perturbations such as obstacles or unexpected changes in visual cues. In order to evaluate this hypothesis the investigators will develop computational and robotic models while systematically performing experiments in rat in which the dorsal or ventral portions of MEC or hippocampus will be inactivated. They will introduce new types of mazes in which the spatial frequency of the trajectories will be controlled. This work will contribute to better spatial navigation in robotics by: (1) providing a robotic testbed to evaluate hypotheses on the role of the entorhinal cortex and (2) providing biologically plausible models for robust spatial navigation under uncertain and dynamic environments. These models will suggest alternatives to classical probabilistic methods commonly used in robot Simultaneous Localization And Mapping paradigms. This work will also contribute to studies of spatial navigation in rats by: (1) showing the usefulness of robots in providing a physical testbed beyond pure computational modeling, and (2) exploiting the shorter cycle of robot experimentation to produce maze configurations that are optimal for testing specific hypotheses in rat experiments.

Spatial navigation is a complex cognitive process that relies on robust and adaptive mechanisms to relate current and future spatial positions to specific locations in the environment. The goal of this project is to provide a better understanding of spatial navigation by integrating information obtained from experimental studies in rats, computational models, and experiments on robots that will test new hypotheses on how these mechanisms work.

The hippocampus and medial entorhinal cortex (MEC) are major brain regions involved in mammalian spatial navigation. While the role of place cells in the hippocampus has been extensively studied, there are still many open questions on the functional role of MEC grid cells and their interaction with the hippocampal place cells. Of interest to this proposal is the recent finding that grid cells are organized in an orderly fashion along the dorso-ventral axis of the MEC, with dorsal grids being much more tightly spaced than ventral ones. The investigators hypothesize that this multiscale organization endows the navigation system with a coding mechanism that will inherently achieve robustness with respect to external perturbations such as obstacles or unexpected changes in visual cues. In order to evaluate this hypothesis the investigators will develop computational and robotic models while systematically performing experiments in rat in which the dorsal or ventral portions of MEC or hippocampus will be inactivated. They will introduce new types of mazes in which the spatial frequency of the trajectories will be controlled. This work will contribute to better spatial navigation in robotics by: (1) providing a robotic testbed to evaluate hypotheses on the role of the entorhinal cortex and (2) providing biologically plausible models for robust spatial navigation under uncertain and dynamic environments. These models will suggest alternatives to classical probabilistic methods commonly used in robot Simultaneous Localization And Mapping paradigms. This work will also contribute to studies of spatial navigation in rats by: (1) showing the usefulness of robots in providing a physical testbed beyond pure computational modeling, and (2) exploiting the shorter cycle of robot experimentation to produce maze configurations that are optimal for testing specific hypotheses in rat experiments.

Spatial sequence learning is a complex cognitive process that enables animals and humans to reliably navigate between different locations in a specific order. The goal of this project is to provide a better understanding of how spatial sequence navigation is learned and optimized by integrating information obtained from experimental studies in rats with computational models and autonomous mobile robots.

Spatial sequence learning has been shown to involve brain areas including the hippocampus and the prefrontal cortex (PFC). Recent studies in the rat have shown that neurons in these two areas spontaneously re-activate in short sequences. Much attention has been paid to reactivation during sleep in the context of long-term memory consolidation. The focus of this project is on the role of replay during the awake state, as the animal is learning across multiple trials during the same session. The hypothesis is that the generation of these short sequences of activity in hippocampus allows for global spatial sequence learning in the PFC. The proposed work involves the development of an integrated model of the hippocampus-PFC network that is able to form spatial navigation sequences incorporating: 1) a replay-driven model for memory formation in the hippocampus and 2) a model of spatial sequence learning in the PFC that uses what is known as reservoir computing. The PFC reservoir will consist of large pools of interconnected neural elements that process information dynamically through reverberations. It will consolidate hippocampal replay sequences into larger spatial sequences that may be later recalled by subsets of the original sequences. The proposed work is expected to generate a new mechanistic understanding of the role of replay in memory acquisition in complex tasks such as sequence learning. That understanding will be leveraged and tested on robotic platforms. Original contributions of the proposed work include 1) the use of hippocampal replay to create small chunks of valid trajectories, 2) the use of reservoir computing to learn spatial sequences using the outputs of the hippocampus model, 3) constraining and testing of the model using electrophysiological data in behaving rats and 4) the use of the resulting model in the embodied-cognitive framework of a robot.

A companion project is being funded by the French National Research Agency (ANR).

Navigation technologies are an increasingly important component of everyday life in an ever more dynamic and complex world. One limitation of these technologies is that they are optimized for a specific spatial scale. Another limitation is that they do not use the knowledge of previous navigation to compute new paths, essentially starting from scratch every time they are invoked. Recent evidence shows, however, that the mammalian brain has evolved to use multiple spatial navigation scales in parallel, and to use spatial memory to improve path planning. How these scales are used and what advantages such uses provide are still unknown. This project hypothesizes that multiscale spatial navigation is crucial in large and cluttered environments. Experiments recording from the neurons of the "brain GPS" system of the rodent (an excellent and efficient spatial navigator) seek to elucidate the basic principles of memory-based multiscale spatial navigation. These experiments will inform new algorithms that will be implemented on a computer to simulate complex multiscale spatial navigation, mimicking the neural computations of the brain. The simulations will then be tested and improved on actual autonomous mobile robots navigating in challenging complex environments.

The project will use wireless high density neural recording technologies allowing for parallel recording of large populations of individual neurons. Optogenetic techniques will be used to manipulate the activity of these neurons and study their impact on the behavior and spatial memory of the animal. The multiscale pattern of neural activity will be used in the development of a mechanistic computational model, which will be tested in new and arbitrary simulated environments, and generate predictions as to how the neural system might succeed or fail. Finally, the simulations will be ported onto a mobile robot, where the algorithms can be tested and improved when the robot is faced with real sensor noise and unreliable world features.