Blanc SIMI 2 - Sciences de l'information, de la matière et de l'ingénierie : Sciences de l’information, simulation

Coding of information on different time scales for spatial decision – NEUROBOT

Electrophysiological unit activity was recorded in rats performing several behavioral tasks involving representations of recent and imminent events at different time scales. The CIRB team characterized representations in the hippocampus (HS), prefrontal cortex (PfC) and striatum (Str) that encode recent and imminent trajectories as well as other responses to learning new rules based upon the outcomes of previous trials via bootstrap Monte Carlo analyses. The LNC team analyzed hippocampal and PfC activity in rats performing a spatial memory task requiring memorization of a series of events for decision making on a longer time scale (minutes). The results arising from these experiments were applied in building artificial neural networks with spiking neurons or firing rate neurons. These models bring new insights into cortical-striatal interactions on the one hand and the cortical-hippocampal-basal ganglia circuit on the other.

Hippocampal retrospective modulation prevails over prospective. PfC neuronal activity underlies tagging of events in working memory for retrospective or prospective recall on the order of minutes. The neuro-computational models reproduced the experimental results. Modelling also allowed us to generate architectures that account for the formation and the articulation between retrospective and prospective memory as found in the experimental work.

-

15 articles in peer-reviewed international journals, 4 book chapters, 29 communications in international conferences, 4 communications in national conferences and 2 broader articles were published. The partners also participated in various public events like la « Nuit des Chercheurs », « Futur en Seine », la « Fête de la Science », la « Semaine du Cerveau » et la « Journée de l’Intelligence Embarquée ». Other publications and presentations are in preparation.

Submission summary

Tracking, memorizing and recalling real world events and episodes poses a formidable problem for artificial as well as biological computational processors. A major issue is time: the underlying mechanisms are much too rapid – nanoseconds and milliseconds for in silico and in vivo systems respectively. Thus efficient solutions must be found to scale up to the timescale of typical experiences which for autonomous robots and animals are on the order of seconds, minutes and hours. Furthermore, optimal operation at these different timescales requires hierarchies of nested systems in order to permit expression of adaptive behaviour at the highest level. Here we will study the nature of this nesting and the control systems for optimal communication between these levels.

For bio-inspired control systems, it is necessary to understand the neural mechanisms underlying cognitive functions. For this, we record brain activity in rats as they perform learning, memory and decision tasks. In particular, maze tasks provide an insight into the rat’s knowledge about the environment and its previous experience. While this approach can bring to light candidate mechanisms by which neural circuits may enact cognitive functions, it is limited in that it only provides correlational, but not causal, evidence. In order to demonstrate whether these algorithms are necessary and sufficient to enable sophisticated behaviours, we replicate the neural principles in artificial systems by computational models and confirm their viability by implementation in mobile artefacts (i.e., robots). This permits us to distinguish processes that are essential for cognitive functions from those that serve only to facilitate implementation in brain tissue. Hence we can create autonomous robots endowed with adaptive and robust control systems thanks to their biological inspiration.

One goal here is to help better understand and efficiently replicate how the brain memorizes and organizes information at several different timescales and how multiple brain areas are coordinated for this.
The regimes focussed on here include:
1.Linking immediately previous events with imminent behavioural choices to form a representation of series of events (on the order of 1-10 s)
2.Linking series of events to make up episodes (on the order of tens of seconds to minutes)
3.Engaging experience from multiple episodes (on the order of hours or more) for decision making (i.e. action selection and strategy selection).

Models for the three regimes have been identified in the hippocampus-prefrontal cortex-striatum (Hpc-Pfc-Str) brain system and will be further highlighted below.
1. Retrospective, prospective activity in Hpc; predictive activity in Pfc and striatum
2. Sequence and memory-requirement dependent activity in Pfc
3. Synchronization of oscillations among brain areas to distinguish a functionally active pathway (and hence behavioural choice)

The teams will participate interactively in studies at each of the time regimes. Furthermore, in order to better understand the links between these levels, each task in the experimental and computational work will span at least two regimes. This will permit us to determine the algorithms carried out, and then to adapt them to develop efficient computational equivalents to embed in higher hierarchical levels.

Project coordination

Mathias QUOY (UNIVERSITE DE CERGY-PONTOISE) – quoy@ensea.fr

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.

Partner

GNT ECOLE NORMALE SUPERIEURE
LNC CNRS - DELEGATION REGIONALE PROVENCE ET CORSE
ETIS UNIVERSITE DE CERGY-PONTOISE

Help of the ANR 829,256 euros
Beginning and duration of the scientific project: - 48 Months

Useful links

Explorez notre base de projets financés

 

 

ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter