SE PhD Final Defense: Zexin Sun

SE PhD Final Defense: Zexin Sun

TITLE: Control and Learning in Bio-Inspired Networks

ADVISOR: John Baillieul SE, ME, ECE

COMMITTEE: Chair: David Castanon (ECE, SE); Ioannis Paschalidis ECE, SE, BME; Sean Andersson ME, SE; Roberto Tron ME, SE

ABSTRACT: The voluntary motions of animals are typically initiated by high level intentions created in the primary cortex through a combination of perceptions of the current state of the environment along with memories of past reactions to similar states. Muscle movements are produced as a result of neural processes in which the parallel activity of large multiplicities of neurons generate signals that collectively lead to desired actions. Inspired by this, we propose \emph{neuromimetic systems} and study models in which cascaded neural activity is the basis of biological motor control. The study involves models at two different scales: a micro scale in which the focus is on parallel activity of large sets of simple inputs that play a role that is analogous to sets of neurons in biological systems, and a macro scale in which different regions of network models interoperate and display characteristic features observed in neurobiology. In the micro scale, we consider overcomplete linear systems that are characterized by larger numbers of input and output channels than the dimensions of the state. A resilient controller and an observer are designed whose operations tolerate input and output channel intermittency and even complete dropouts. We also propose a principled quantization in which control signals are encoded as simple discrete inputs which act collectively through the channels of input that are the hallmarks of the overcomplete models. Aligned with the neuromimetic paradigm, an emulation problem is proposed and this in turn defines an optimal quantization problem. Several possible approaches are discussed including direct combinatorial optimization, a Hebbian-like iterative learning algorithm, and a deep Q-learning (DQN) approach. For the problems being considered, machine learning approaches to optimization provide valuable insights regarding comparisons between optimal and nearby suboptimal solutions. We also make an extension of the emulation problem to a nonlinear system by applying the Koopman Operator with the deep deterministic policy gradient (DDPG) algorithm. In the macro scale, we propose a biologically inspired network model featuring dynamic connections regulated by Hebbian learning. Drawing on the machinery of graph theory and classical control we show that our novel nonlinear model exhibits such biologically plausible features as bounded evolution, stability, resilience, and a kind of structural stability -- meaning that perturbations of the model parameters leave the essential properties of the model in tact. The proposed network model involves generalized cactus graphs with multiple control input nodes, and it is shown that the properties of the network are resilient to various changes in network topology provided these changes preserve the generalized cactus structure. A particular example described in what follows is an idealized network model of the visual system of a macaque monkey. The model displays resilience to network disruptions such as might occur in a living organism due to disease or brain injury.