Directory

Synaptic Plasticity

Like many biological processes, synaptic plasticity requires a compensatory mechanism, commonly referred to as synaptic homeostasis, to help ensure that the nervous system is in a dynamic regime where it functions optimally. Synaptic homeostasis also changes and adapts to the activity of the brain. This process is called homeostatic plasticity. A review of the literature reveals a varied and somewhat paradoxical set of findings about homeostatic plasticity. While in most theoretical models, it needs to be fast -- in seconds or minutes -- and sometimes even instantaneous to achieve stability, experimentalists witness slower homeostatic plasticity in the order of hours or days. It has, in fact, been suggested that both fast and slow homeostatic mechanisms exist and that learning and memory use an interplay of both forms of homeostasis: while fast homeostatic control mechanisms (also referred to as rapid compensatory processes or RCPs) help maintain the stability of synaptic plasticity, slower ones are important for fine-tuning neural circuits.

Dr. Udeigwe’s DoD funded work uses analytical, numerical, and data-driven machine learning methods to investigate and model the dynamical interactions between synaptic plasticity and homeostatic plasticity. The overall goal is made up of three specific aims (1) to develop and analyze a multistable learning rule, with multiple firing rate setpoints, that incorporates a slow homeostasis mechanism as well two RCPs to model learning and memory formation; (2) to analyze the dynamics of -- and the possible interaction between -- synaptic plasticity and homeostatic plasticity during recurrent network activities; (3) to develop a data-driven theoretical framework in which the nonlinear evolution of synaptic activity dynamics can be approximated with a linear operator, with the secondary goal of extracting important timescale information. Investigation is done at both single-neuron and network levels and will employ a combination of theoretical tools including -- but not limited to -- stability and bifurcation analysis, perturbation analysis, data-driven spectral decomposition and deep learning of recurrent artificial recurrent neural networks.

Visual Object Recognition

By using electrophysiological recordings in animal models, it has been observed that, in all primary sensory brain regions, the firing rate of individual neurons reduces in response to a stimulus that is enlarged beyond a spatial region -- known as receptive field -- associated with these neurons, in which occupying stimuli are able to induce neuronal activity. This phenomenon is known as surround suppression, and has been studied for neurons in the primary visual cortex (V1), primary somatosensory (touch) cortex (S1), primary auditory cortex (A1), and other parts of the visual pathway . Surround suppression has been associated with a range of cognitive functions including contour integration (detecting continuity of curved and/or broken edges), motion detection, velocity detection, and texture segregation.

Dr Udeigwe’s work, in collaboration with the DiCarlo Lab at BCS, uses artificial neural networks (ANN) to investigate and further understand the mechanisms behind surround suppression and other vision normalization processes. The central idea is that, by building networks that can accomplish visual intelligence tasks, we can ask about the importance of surround suppression (a local phenomenon) in terms of overall network performance (a global phenomenon).