Neural Coding and Feature Extraction of Time-Varying Signals
What are the neuronal codes that the brain uses to represent information? This constitutes one of the most fascinating and challenging questions in Neuroscience. Here we report the results of our investigations about the mechanisms of stimulus encoding and feature extraction using the weakly electri...
| Main Author: | |
|---|---|
| Format: | Others |
| Language: | en |
| Published: |
2002
|
| Online Access: | https://thesis.library.caltech.edu/2076/1/kreiman_thesis2_wfigs.pdf Kreiman, Gabriel Alejandro (2002) Neural Coding and Feature Extraction of Time-Varying Signals. Master's thesis, California Institute of Technology. doi:10.7907/2V9Y-A983. https://resolver.caltech.edu/CaltechETD:etd-05262002-173148 <https://resolver.caltech.edu/CaltechETD:etd-05262002-173148> |
| Summary: | What are the neuronal codes that the brain uses to represent information? This constitutes one of the most fascinating and challenging questions in Neuroscience. Here we report the results of our investigations about the mechanisms of stimulus encoding and feature extraction using the weakly electric fish Eigenmannia as a model. In many circumstances, sensory systems are subject to natural stimuli that are constantly changing. Therefore we decided to study the representation of time varying signals. Eigenmannia constitutes an ideal system to combine neurophysiological and computational techniques to study neural coding. We have characterized the variability of neuronal responses with a new approach by using parameterized distances between spike trains defined by Victor and Purpura. This measure of variability is widely applicable to neuronal responses, irrespective of the type of stimuli used (deterministic versus random) or the reliability of the recorded spike trains. We also quantitatively defined and evaluated the robustness of the neural code to spike time jittering, spike failures and spontaneous spikes. Our data show that the intrinsic variability of single spike trains lies outside of the range where it might degrade the information conveyed, yet still allows for improvement in coding by averaging across multiple afferent fibers. We also built a phenomenological model of P-receptor afferents incorporating both their linear transfer properties and the variability of their spike trains. We then studied the extraction of features from the time varying signal by bursts of spikes of multiple pyramidal cells, the next stage of information processing. To address the question of whether correlated responses of nearby neurons within topographic sensory maps are merely a sign ofredundancy or carry additional information we recorded simultaneously from pairs of electrosensory pyramidal cells with overlapping receptive fields in the hindbrain of weakly electric fish. We found that nearby pyramidal cells exhibit strong stimulus-induced correlations. The detailed stimulus encoding by pairs of pyramidal cells was inferior to that from single primary afferents. However, the detection of coincident bursts of activity could significantly enhance the extraction of upstrokes and downstrokes in the stimulus amplitude. Our investigations reveal mechanisms by which the nervous system can accurately and robustly transduce a time-varying signal into a digital spike train and then extract behaviorally relevant features. |
|---|