Cognition, behavior and systems neuroscience

PROGRAMME

9h20-9h30 M VERGASSOLA : Introduction and a few words on QBio.
9h30-10h Clément LENA
10h-10h30 Ralitsa TODOROVA
Pause
11h-11h30 Chistoph SCHMIDT-HIEBER
11h30-12h Alex CAYCO-GAYJIC.
12h-12h30 Rava AZEREDO da SILVEIRA
14h-14h30 Gisella VETERE
14h30-15h Jean-Rémi KING
15h30-16h Auriane DUCHEMIN
16h-16h30 Philippe FAURE

TITLES AND ABSTRACT

Rava AZEREDO DA SILVEIRA (LPENS)
Cognitive Biases and Compressed Mental Representations

Alex CAYCO-GAJIC (DEC-ENS)
High-dimensional neural representations during sensorimotor behaviour

Auriane DUCHEMIN (IBENS)
Neural circuit dynamics underlying switches in visual perception in the zebrafish larva
Usually the sensory information is non ambiguous and there is only one correct interpretation of the stimulus. But sometimes the same stimulus can be perceived in different ways and this leads to alternation of the interpretation separated by switches in the perception. To shed light on the circuit processes and mechanisms underlying this phenomenon, I am studying the zebrafish larva which displays a robust and stereotyped behavior, the optokinetic response (OKR, innate eyes motion), when presented with a moving vertical grating. I created a modified version of this stimulus that, when displayed continuously to the larva, leads to periods of perception of movement and periods of spontaneous eye rotations that suggest no perception of the movement. I am investigating how these periods are represented in the zebrafish brain, and also what are the neuronal circuit dynamics that lead to the switch between both, using two-photon calcium imaging and light sheet microscopy in zebrafish larvae expressing GCaMP6f.

Philippe FAURE (Plasticité du Cerveau, ESPCI)
Exploratory behavior, individual traits and vulnerability to drug

Jean-Rémi KING (LSP, DEC-ENS)
Language Processing in Brains and in Deep Neural Networks: Computational Convergence and its Limits
Deep learning has recently allowed substantial progress in language tasks such as translation and completion. Do these models process language similarly to humans, and is this similarity driven by systematic structural, functional and/or learning principles? To address these issues, we test whether the activations of artificial neural networks trained on (1) image, (2) word and (3) sentence processing map onto human brain responses to written words, as recorded with magneto-encephalography (MEG) and functional Magnetic Resonance Imaging (fMRI) in 202 healthy subjects, and 176 patients implanted with intracranial electrodes. Our results reveal that visual, word and language models respectively correlate with distinct areas and dynamics of the left-lateralized cortical hierarchy of reading. However, only specific subsets of these models converge towards brain-like representations during their training. Specifically, the middle layers of language models become increasingly similar to the late responses of associative cortices. By contrast, training algorithms lead them to generate input and output representations that differ from those of the brain. These differences allowed us to precisely track and dissociate the perceptual, lexical and compositional representations generated during reading in each cortical region. Overall, our results suggest that only specific subsets of modern language models converge to the language representations elicited in by the human brain.

Clément LENA (IBENS)
A distributed network for motor learning in the mouse
Three main brain structures are often found in motor learning: the cortex, the striatum and the cerebellum. Each of these structures have unique computational abilities and show variation in their involvement in motor tasks during learning. Here we investigated the contribution to learning and execution of the connections of the cerebellum with the thalamo-cortical and thalamo-striatal system in motor learning. Our findings demonstrate a dual contribution of the cerebellar interactions with these system along learning, and notably reveal an offline process converting savings into consolidated memories.

Christoph SCHMIDT-HIEBER (Institut Pasteur)
Novelty drives a switch from neuronal generalization to discrimination in the hippocampus
Memory encoding and retrieval are complementary processes that put conflicting requirements on neuronal computations in the hippocampus. How this challenge is resolved in hippocampal circuits is unclear. To address this question, we combine electrophysiology, imaging, and computational modelling. We find that neurons in the dentate gyrus consistently show a small transient depolarization upon transition to a novel environment. A computational model suggests that the observed transient synaptic response to novel environments can lead to a bias in the granule cell population activity, which in turn drives the downstream CA3 attractor network to a new state, thereby favoring the switch from generalization to discrimination when faced with novelty. This switch enables encoding of novel memories while preserving stable retrieval of familiar ones. 

Ralitsa TODOROVA (Collège de France)
Isolated cortical computations during delta waves
The hippocampus and the neocortex communicate in sleep to consolidate memories. Neuronal "replay" in the hippocampus tends to be followed by cortical delta waves - short moments of cortical silence. How can this silence help consolidate the information just received? It turns out that not all neurons remain silent during delta waves, and the few neurons that remain actually perform important computations.

Gisella VETERE (LSP, Brain Plasticity lab, ESPCI)
Decoding memory formation and stabilization in mice
How the brain processes information from the world outside us to save it in the neural network? Can we implant false memories bypassing the external experience via artificial manipulation of known neuronal pattern? In this talk I will show that, by simply knowing the identity of the cells that are responsible for encoding a specific external stimulus, we can create a memory of an event never experienced before. To form a long-lasting memory trace, the brain undergoes a substantial rearrangement at both synaptic and systems level to support the stabilization and future recall of the experienced event. What are the regions involved in this stabilization process? I will explore a graph theory analysis of memory networks to show that this approach opens new doors in the discovery of brain regions and pathways involved in cognitive functions. Finally, I will present the research interests of the C4 (Cerebral Codes and Circuits Connectivity) team that I am leading at the ESPCI in Paris: we are studying how neuronal codes used to process spatial information are modified during memory stabilization. To this aim, we take advantage of cutting-edge techniques including the use of miniaturized microscopes to detect calcium activity in deep brain regions in freely moving mice and optogenetics to manipulate target neuronal circuits and causally link their role to learning and memory processes.