[month] [year]

Indo-French SARASWATI research project 

A 3-year, DST-INRIA, Indo-French project – SARASWATI: Sequential motor skills: a dual system view has been started with Prof. Bapi Raju Surampudi, Cognitive Science Lab, and Nicolas P. Rougier, INRIA Bordeaux Sud-Ouest  as the principal investigators   and Srinivasa Chakravarthy, Department of Biotechnology, IIT Madras as the Co-principal investigator.

 The objectives for the three years is to design a generic machine learning architecture mixing Hebbian learning and reinforcement learning and to compare this new architecture to more classical approaches (supervised learning). Based on experimental evidence on the French side (newts, rodents and non-human primates) as well as behavioral investigation on humans (IIITH), the two teams will explore computational models that can give account on behavior and will also test their respective hypotheses. We’ll therefore propose a theory of motor learning and skill acquisition in which the basal ganglia leads learning in the early stage, and progressively passes on the results of its learning in the later stages to the motor cortex. In the early stages of learning, the basal ganglia combines the sensory-motor information and the reward/outcome information, and models the value function. Movement in the early stage is driven by the slow hill-climbing dynamics over the Value function modeled by the basal ganglia (striatum). Thus in the early stages the agent discovers the optimal action by the slow process of hill-climbing of the value function. In the later stages, the optimal actions discovered by the basal ganglia are passed on to the sensory-motor cortical pathway, which performs actions rapidly in a stimulus-response fashion. These models will be tested and evaluated using a multi-segment arms with a various number of segments such as to make the reward function more or less sparse. A further development of the above modeling approach would involve modeling the Prefrontal Cortex and the Basal Ganglia as a hierarchical Reinforcement Learning (RL) System. In such a hierarchical RL framework, the PFC acts as a higher level agent, with the basal ganglia and cerebellum acting as lower level agents. It is then possible to have a more flexible framework for describing motor learning, since the exact agent that leads or follows motor learning in such a framework, could flexibly depend on the task conditions and the context.

One important objective of the associate team will be to train young researchers to these kinds of neural networks and to their exploitation in an AI framework. Reconciling the problem solving and learning sides of AI is a major aim in research today and the associate team, if successful, will gain a high visibility allowing for publications with high impact.