Thesis (M.Phil.)--Hong Kong University of Science and Technology, 2008. / Includes bibliographical references (leaves 74-75). Also available in electronic version.

Thesis (M.S.)--Rochester Institute of Technology, 2009. / Typescript. Includes bibliographical references (leaves 145-149).

Thesis (Ph. D.)--University of Hong Kong, 2005. / Title proper from title frame. Also available in printed format.

Thesis (M. A.)--University of North Texas, May, 2007. / Title from title page display. Includes bibliographical references.

A growing body of research indicates that much of the brain’s computation is invisible from the activity of individual neurons, but instead instantiated via population-level dynamics. According to this ‘dynamical systems hypothesis’, population-level neural activity evolves according to underlying dynamics that are shaped by network connectivity. While these dynamics are not directly observable in empirical data, they can be inferred by studying the structure of population trajectories. Quantification of this structure, the ‘trajectory geometry’, can then guide thinking on the underlying computation. Alternatively, modeling neural populations as dynamical systems can predict trajectory geometries appropriate for particular tasks. This approach of characterizing and interpreting trajectory geometry is providing new insights in many cortical areas, including regions involved in motor control and areas that mediate cognitive processes such as decision-making. In this thesis, I advance the characterization of population structure by introducing hypothesis-guided metrics for the quantification of trajectory geometry. These metrics, trajectory tangling in primary motor cortex and trajectory divergence in the Supplementary Motor Area, abstract away from task-specific solutions and toward underlying computations and network constraints that drive trajectory geometry. Primate motor cortex (M1) projects to spinal interneurons and motoneurons, suggesting that motor cortex activity may be dominated by muscle-like commands. Observations during reaching lend support to this view, but evidence remains ambiguous and much debated. To provide a different perspective, we employed a novel behavioral paradigm that facilitates comparison between time-evolving neural and muscle activity. We found that single motor cortex neurons displayed many muscle-like properties, but the structure of population activity was not muscle-like. Unlike muscle activity, neural activity was structured to avoid ‘trajectory tangling’: moments where similar activity patterns led to dissimilar future patterns. Avoidance of trajectory tangling was present across tasks and species. Network models revealed a potential reason for this consistent feature: low trajectory tangling confers noise robustness. We were able to predict motor cortex activity from muscle activity by leveraging the hypothesis that muscle-like commands are embedded in additional structure that yields low trajectory tangling. The Supplementary Motor Area (SMA) has been implicated in many higher-order aspects of motor control. Previous studies have demonstrated that SMA might track motor context. We propose that this computation necessitates that neural activity avoids ‘trajectory divergence’: moments where two similar neural states become dissimilar in the future. Indeed, we found that population activity in SMA, but not in M1, reliably avoided trajectory divergence, resulting in fundamentally different geometries: cyclical in M1 and helix-like in SMA. Analogous structure emerged in artificial networks trained without versus with context-related inputs. These findings reveal that the geometries of population activity in SMA and M1 are fundamentally different, with direct implications regarding what computations can be performed by each area. The characterization and statistical analysis of trajectory geometry promises to advance our understanding of neural network function by providing interpretable, cohesive explanations for observed population structure. Commonality between individuals and networks can be uncovered and more generic, task-invariant, fundamental aspects of neural response can be explored.

Artificial recurrent neural networks (RNNs) are powerful models for understanding and modeling dynamic computation in neural circuits. As such, RNNs that have been constructed to perform tasks analogous to typical behaviors studied in systems neuroscience are useful tools for understanding the biophysical mechanisms that mediate those behaviors. There has been significant progress in recent years developing gradient-based learning methods to construct RNNs. However, the majority of this progress has been restricted to network models that transmit information through continuous state variables since these methods require the input-output function of individual neuronal units to be differentiable. Overwhelmingly, biological neurons transmit information by discrete action potentials. Spiking model neurons are not differentiable and thus gradient-based methods for training neural networks cannot be applied to them. This work focuses on the development of supervised learning methods for RNNs that do not require the computation of derivatives. Because the methods we develop do not rely on the differentiability of the neural units, we can use them to construct realistic RNNs of spiking model neurons that perform a variety of benchmark tasks, and also to build networks trained directly from experimental data. Surprisingly, spiking networks trained with these non-gradient methods do not require significantly more neural units to perform tasks than their continuous-variable model counterparts. The crux of the method draws a direct correspondence between the dynamical variables of more abstract continuous-variable RNNs and spiking network models. The relationship between these two commonly used model classes has historically been unclear and, by resolving many of these issues, we offer a perspective on the appropriate use and interpretation of continuous-variable models as they relate to understanding network computation in biological neural circuits. Although the main advantage of these methods is their ability to construct realistic spiking network models, they can equally well be applied to continuous-variable network models. An example is the construction of continuous-variable RNNs that perform tasks for which they provide performance and computational cost competitive with those of traditional methods that compute derivatives and outperform previous non-gradient-based network training approaches. Collectively, this thesis presents efficient methods for constructing realistic neural network models that can be used to understand computation in biological neural networks and provides a unified perspective on how the dynamic quantities in these models relate to each other and to quantities that can be observed and extracted from experimental recordings of neurons.

Thesis (M.S.)--Rochester Institute of Technology, 2009. / Typescript. Includes bibliographical references (p. 120-126).

Thesis (Ph. D.)--Rutgers University, 2009. / "Graduate Program in Biology." Includes bibliographical references (p. 107-114).

Thesis (Ph. D.)--University of California, San Diego, 2001. / Vita. Includes bibliographical references (leaves 115-120).

published_or_final_version / abstract / Physics / Doctoral / Doctor of Philosophy