Creating the synthetic brain through hybrid computational and biological systems repairing and replacing neural networks

TitleCreating the synthetic brain through hybrid computational and biological systems repairing and replacing neural networks
Publication TypeConference Paper
Year of Publication2010
AuthorsFrancis, J. T., Chapin J., Lytton WW., Barbour R., Carmena J., Principe J., Sanchez J., & Fortes J.
Conference NameSociety for Neuroscience 2010 (SFN '10)
KeywordsSFN, Society for Neuroscience
Abstract

Here we introduce a joint effort between eight laboratories with the goal of producing an InSilico brain to be used in conjunction with a brain machine interface. Unique to this work is our use of reinforcement learning (RL) as a guiding principal. We will situate this RL appropriately in a multilaminar neocortical model to be used in our brain machine interface (BMI). In addition, a computational agent that learns via RL will be placed between the user, the InSilico brain and the environment. A goal of this work will be to produce a BMI that will allow users to make dexterous reaching and grasping movements with seamless control. This system will give the user somatosensory feedback to help in such control. This system will also be able to compensate for degradation in neural recordings over time. In addition, our system incorporates a computational agent that learns via RL to compensate for any loss in the user's ability such as that due to a damaged cortical region. We will use large-scale computer simulations based on the real architecture of the nonhuman primate sensorimotor thalamocortical systems. Three primate labs will work in collaboration to develop and instrumented environment in which macaque monkeys will make reach to grasp and transport movements. Within this environment all kinematic and dynamic variables will be controlled by the experimenter. As such, the weights, shapes and dynamic properties of a set of virtual objects will be seamlessly controllable at the experimenters will. The monkeys will learn to grasp these virtual objects and transport them in the proper orientation to place them into a virtual pegboard. Large-scale microelectrode arrays will be placed along the sensorimotor neural stream in addition to EEG and fNIRS. The brain regions to be implanted are the primary sensory and motor cortices including areas 3a, 3b, 1 and 2. In addition, recordings will be taken from PMD, PMV and the parietal reach region (PPR). A set of deep brain structures will also be recorded from including the ventral posterior lateral thalamus, the nucleus accumbens and the striatum. Preliminary results from this large-scale collaborative effort will be presented.