Fairhall, Adrienne LPang, Rich2019-05-022019-05-022019-05-022019Pang_washington_0250E_19673.pdfhttp://hdl.handle.net/1773/43738Thesis (Ph.D.)--University of Washington, 2019A central goal in theoretical neuroscience is to understand how neural systems perform computations over the continuum of timescales that underlie behavior. In particular, what are the algorithms and mechanisms enabling single-neuron membrane voltage fluctuations, which occur over milliseconds, to produce the dynamics and information processing in behavior that unfold over hours to years? Notably, while the core ionic processes of membrane voltage fluctuations have been largely elucidated and while extensive theories and evidence exist to explain how slow modulation of neural network structures might underlie learning, almost nothing is known about the liminal regime of seconds to minutes that bridges these two timescales. In the work that follows I address three questions in three different systems, each of which centers around neural computations occurring over the timescales of seconds to minutes. I first investigate the navigational decisions made by flying insects during odor tracking, where I show that fruit flies and mosquitoes exhibit a history dependence in their odor-triggered turning responses that is qualitatively similar to an information-maximizing tracking strategy, but not to others. Next, in collaboration with Ari Zolin, Raphael Cohn, and Vanessa Ruta, I analyze the dynamics of dopaminergic neuromodulation of a short-term memory circuit in the fruit fly mushroom body, where we suggest that the fly dopamine system encodes multiplexed representations of a wide diversity of sensory, motor, and valence signals, some of which predict behavior several seconds in the future. Third, I develop a spiking neural network model capable of storing and replaying sequential activity patterns using a heterosynaptic and fast-acting biological plasticity rule, and which reconstructs sequences through the existing recurrent network structure. Collectively, these results elucidate the computational capacities of three distinct systems and shed new light on short-term information processing in neural computations from three novel angles. Finally, in collaboration with Sid Henriksen and Mark Wronkiewicz, I describe a simple network-growth model reproducing several statistical features of mouse brain network connectivity at the mesoscale; while this work does not explicitly address short-term computations, simplified statistical network models will be crucial to eventually understanding how such computations occur within large scale distributed brain networks.application/pdfen-USCC BYcomputationdopaminememorynetworkturbulenceNeurosciencesApplied mathematicsPhysiology and biophysicsThesis