The study of learning has been a central focus of psychology and neuroscience since their inception. Cognitive neuroscience’s traditional approach to understanding learn-ing has been to decompose it into discrete cognitive processes with separable and localized underlying neural systems. While this focus on modular cognitive functions for individual brain areas has led to considerable progress, there is increasing evidence that much of learn-ing behavior relies on overlapping cognitive and neural systems, which may be harder to disentangle than previously envisioned. This is not surprising, as the processes underlying learning must involve widespread integration of information from sensory, affective, and motor sources. The standard tools of cognitive neuroscience limit our ability to describe processes that rely on widespread coordination of brain activity. To understand learning, it will be necessary to characterize dynamic co-activation at the circuit level.
In this dissertation, I present three studies that seek to describe the roles of distrib-uted brain networks in learning. I begin by giving an overview of our current understand-ing of multiple forms of learning, describing the neural and computational mechanisms thought to underlie incremental feedback-based learning and flexible episodic memory. I will focus in particular on the difficulties in separating these processes at the cognitive level and in localizing them to individual regions at the neural level. I will then describe recent findings that have begun to characterize the brain’s large-scale network structure, emphasiz-ing the potential roles that distributed networks could play in understanding learning and cognition more generally. I will end the introduction by reviewing current attempts to char-acterize the dynamics of large-scale brain networks, which will be essential for providing a mechanistic link to learning behavior.
Chapter 2 is a study demonstrating that intrinsic connectivity between the hippo-campus and the ventromedial prefrontal cortex, as well as between these regions and dis-tributed brain networks, is related to individual differences in the transfer of learning on a sensory preconditioning task. The hippocampus and ventromedial prefrontal cortex have both been shown to be involved in this type of learning, and this study represents an early attempt to link connectivity between individual regions and broader networks to learning processes.
Chapter 3 is a study that takes advantage of recent developments in mathematical modeling of temporal networks to demonstrate a relationship between large-scale network dynamics and reinforcement learning within individuals. This study shows that the flexibil-ity of network connectivity in the striatum is related to learning performance over time, as well as to individual differences in parameters estimated from computational models of re-inforcement learning. Notably, connectivity between the striatum and visual as well as or-bitofrontal regions increased over the course of the task, which is consistent with an inte-grative role for the region in learning value-based associations. Network flexibility in a dis-tinct set of regions is associated with episodic memory for object images presented during the learning task.
Chapter 4 examines the role of dopamine, a neurotransmitter strongly linked to val-ue updating in reinforcement learning, in the dynamic network changes occurring during learning. Patients with Parkinson’s disease, who experience a loss of dopaminergic neu-rons in the substantia nigra, performed a reversal-learning task while undergoing functional magnetic resonance imaging. Patients were scanned on and off of a dopamine precursor medication (levodopa) in a within-subject design in order to examine the impact of dopa-mine on brain network dynamics during learning. The reversal provided an experimental manipulation of dynamic connectivity, and patients on medication showed greater modula-tion of striatal-cortical connectivity. Similar results were found in a number of regions re-ceiving midbrain projections including the prefrontal cortex and medial temporal lobe. This study indicates that dopamine inputs from the midbrain modulate large-scale network dy-namics during learning, providing a direct link between reinforcement learning theories of value updating and network neuroscience accounts of dynamic connectivity.
Together, these results indicate that large-scale networks play a critical role in multi-ple forms of learning behavior. Each highlights the potential importance of understanding dynamic routing and integration of information across large-scale circuits for our concep-tion of learning and other cognitive processes. Understanding the when, where, and how of this information flow in the brain may provide an alternative or compliment to traditional theories of distinct learning systems. These studies also illustrate challenges in integrating this perspective with established theories in cognitive neuroscience. Chapter 5 will situate the studies in a broader discussion of how brain activity relates to cognition in general, while pointing out current roadblocks and potential ways forward for a cognitive network neuroscience of learning.
Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/D84J1Z0R |
Date | January 2018 |
Creators | Gerraty, Raphael Thomas |
Source Sets | Columbia University |
Language | English |
Detected Language | English |
Type | Theses |
Page generated in 0.0557 seconds