UC Irvine

This dissertation explores the use of computational models, specifically spiking neural network (SNN) models combined with evolutionary computation, as a versatile approach to investigate neural representations and underlying neural mechanisms involved in navigation. The aim of this thesis is to understand how information is processed and represented in the cortical motion stream for visual motion perception and in the hippocampus for spatial cognition.

The first study introduces CARLsim, an efficient SNN simulator with biologically detailed implementation of spiking neurons, synapses, and various synaptic plasticity rules. Additionally, CARLsim is integrated with evolutionary computation libraries, providing flexible and powerful modeling capabilities. This tool is utilized in the modeling studies presented in this dissertation.

In the second study, an SNN model is developed to understand the complex response properties of neurons in the dorsal sub-region of the Medial Superior Temporal area (MSTd) area and how they support heading estimation. The resulting model showed receptive fields that are suitable for processing self-motion-induced optic flows, and accurately encoded the heading direction in the neuron population. This study explores how synaptic plasticity rules could implement nonnegative sparse coding (NSC) for efficient coding, and supports a previous theory which suggests that receptive fields observed in MSTd may emerge through dimensionality reduction on its input.

The third study investigates how spatial representations that are important for navigation emerge with self-motion cues and external cues that signal the relationship between the agent and the environment. SNN models of hippocampal sub-region CA1 and the subiculum (SUB) were developed using neuronal data recorded in a working memory navigational task. The models exhibited differential spatial representations that reflect their functional distinctions and match experimental observations. This study shows that distinct representations of spatial features can be formed by different weightings in the integration of navigational variables.

This dissertation provides insights into how the brain processes information during navigation, shedding light on the neural mechanisms underlying visual motion perception and spatial cognition. The presented studies highlight the potential of SNNs and evolutionary computation as powerful tools for computational modeling of the brain, which offer the advantage of biological plausibility and minimal assumptions on the model parameters. This research may inspire the development of other novel modeling approaches. The models developed in this research could have practical applications in robotics and artificial intelligence.