Computational modeling of neural network modifications during the critical period
In the early development of the visual cortex, experimental researchers found a “critical period” where the developing network shows very high plasticity in an activity-dependent manner (Levelt and Hubener, 2012). For instance, ocular dominancy, the relative input from the two eyes (Hubel and Wiesel, 1962), remains plastic in only a short period of time, usually several days to weeks after the birth in mice. It was shown that the innervation of GABAergic inhibitory neurons is critical for the onset of this critical period (Hensch et al., 1998), and there is growing evidence that changes in the properties of inhibitory cells (Parvalbumin expressing cells) and inhibitory plasticity might be responsible for activity-driven developmental changes during this period (see (Levelt and Hubener, 2012) for a review).
Although the biological evidence suggests that there is a critical period of plastic changes which could be related to the theoretical work, there are few studies to date relating theoretical insights of learning directly with the experimental observed neural network dynamics during the critical period. Most theoretical models are instead mostly generative statistical models of learning probabilistic patterns and thus too abstract to be related to temporal activity of biological realistic neurons; in fact, dynamics of neurons is never modeled in these generative models. Thus generative models never mention nor offer any explanation why, e.g., the inhibitory network modification is such important, nor what kind of learning rule for synaptic modifications might be used during this developmental period. Thus there remains a large gap between the experimental observations and the abstract conceptual models. Closing this gap with computational modeling close to experimental observation is the goal of this proposed sub-project.