Download (2.8 MB)
An important question in sensory neuroscience is what coding strategies and mechanisms are used by the brain to detect and discriminate among behaviorally relevant stimuli. To address the noisy response properties of individual neurons, sensory systems often utilize broadly tuned neurons with overlapping receptive fields at the system's periphery, resulting in homogeneous responses among neighboring populations of neurons. It has been hypothesized that progressive response heterogeneity in ascending sensory pathways is evidence of an efficient encoding strategy that minimizes the redundancy of the peripheral neural code and maximizes information throughput for higher level processing. This hypothesis has been partly supported by the documentation of neural heterogeneity in various cortical structures.
This dissertation will examine whether selective and sensitive responses to behaviorally relevant stimuli contribute to a heterogeneous and efficient encoding in the auditory midbrain. Prior to this study, no compelling experimental framework existed to address this question. Stimulus design methodologies for neuroethological experiments were largely based on token vocalizations or simple approximations of vocalization components. This dissertation describes a novel state-space signal modeling methodology which makes possible the independent manipulation of the frequency, amplitude, duration, and harmonic structure of vocalization stimuli. This methodology was used to analyze four mouse vocalizations and create a suite of perturbed variants of each of these vocalizations. Responses of neurons in the mouse inferior colliculus (IC) to the natural vocalizations and their perturbations were characterized using measures of both spike rate and spike timing. In order to compare these responses to those of peripheral auditory neurons, a data-driven model was developed and fit to each IC neuron based on the neuron's pure tone responses. These models were then used to approximate how peripheral auditory neurons would respond to our suite of vocalization stimuli. Using information theoretic measures, this dissertation argues that selectivity and sensitivity by individual neurons results in heterogeneous population responses in the IC and contributes to the efficient encoding of behaviorally relevant vocalizations.
Lars Holmstrom is a Ph.D. candidate in the Systems Science Graduate Program at Portland State University. His research has primarily been focused on reinforcement learning in artificial neural networks and sensory processing in real neural networks. He is currently working as a software architect for a local biomedical device company and as an environmental consultant responsible for model based estimates of avian mortality risk resulting from wind farm installations.
Auditory pathways, Auditory cortex -- Research, Senses and sensation, Auditory perception, System theory -- Applications to neurosciences
Holmstrom, Lars Andreas, "Efficient Encoding of Vocalizations in the Auditory Midbrain" (2010). Systems Science Friday Noon Seminar Series. 34.