The Hybrid Systems laboratory (HLB) is primary involved in the development, implementation and analysis of machine-brain-machine interfaces.
The interdisciplinary nature of the study of hybrid systems lies at the intersection of different research areas, namely:
- computational neuroscience
- artificial intelligence
- neuromorphic engineering
The HLB was involved in the ReNaChip FP7 project, whose overarching goal is to build s neuroprosthetic neuromorphic chip recovering a learning function lost in the aged cerebellum.
The main topics dealt with in this laboratory are currently:
Models of the two phase theory of conditioning: The Polish psychologist Konorski has proposed in the early 1960ies that the associative processes underlying classical conditioning can be separated in a fast non-specifi learning systems (NLS) and a slow specfic one (SLS). We have successfully integrated into a computational model the roles of the amygdala, basal forebrain and auditory cortex (AC) as an example of NLS, and the cerebellum (CE) as a model of SLS. Currently, we are extending the models of AC and CE in order to more robustly match the learning dynamics in observed behavioral experiments. On the robotics side, we are evaluating the learning performance of an extended composite model at the circuit level in simulated conditioning experiments, as well as at the behavioral level using a mobile robot.
We can demonstrate that the model displays both a NLS and a SLS under real-world conditions, and that the learning dynamics of the two subsystems consistently match behavioral observed data. Thus, we can provide a complete account of Konorski’s proposal by integrating these two systems into a physical device and showing that the observed learning dynamics match experimental data.
Study of the hypothesis of plasticity in the cerebellar deep nucleus
A biologically plausible mechanism for the adaptive timing in the granular layer: Decades of experimental work using the conditioning of the eye-blink response paradigm points to the Cerebellar Cortex (CC) as the locus of the motor conditioning engram. It is in the Cerebellar Cortex where a memory trace is formed that allows both to associate Conditioning Stimulus (CS) and Unconditioned Stimulus (US) and to elicit an adaptively timed Conditioned Response (CR). The mechanisms by which this task is accomplished are not yet fully understood but, historically, researchers have postulated that the main substrate for the acquisition of the CR is Long-Term Depression (LTD) at the parallel fiber (PF) to Purkinje cell (PU) synapse. However the assumption that LTD is sufficient for acquiring correctly timed responses requires that information about the time passed since CS onset has to be encoded by the spatio-temporal pattern of activity at the PFs. In our group, we propose that a possibility would be a decreasing memory trace, but the substrate for it remains an open question. Currently, by means of simulations, we are studying how we can obtain an appropriated pattern of spatio-temporal activity based on biologically founded hypotheses.