Directly to content
  1. Publishing |
  2. Search |
  3. Browse |
  4. Recent items rss |
  5. Open Access |
  6. Jur. Issues |
  7. DeutschClear Cookie - decide language by browser settings

Neural Representations and Decoding with Optimized Kernel Density Estimates

Kornienko, Olga

[thumbnail of Thesis_Kornienko2015A.pdf]
Preview
PDF, English
Download (5MB) | Terms of use

Citation of documents: Please do not cite the URL that is displayed in your browser location input, instead use the DOI, URN or the persistent URL below, as we can guarantee their long-time accessibility.

Abstract

In in-vivo neurophysiology, firing rates from single neurons are traditionally presented in the form of spike counts or peri-stimulus time histograms which are accumulated and averaged across many presumably identical trials. These histograms may on the one hand provide either only noisy representations of the true underlying spiking activity, or on the other hand do not enable single trial resolution. Kernel density estimates (KDE), a weighted moving average with Gaussian kernels centered around spike times, act as a low-pass filters averaging out rapid changes in the firing frequency. Optimized KDEs with the width of the Gaussians (bandwidth) determined through cross-validation or bootstrapping reflect more accurately the underlying spiking activity and also allow for single trial resolution. We found that optimized bandwidth estimates obtained through unbiased cross-validation (UCV) are an information rich measure, which is applicable to more problems than firing rate estimation, by analyzing both simulations and multiple single-unit recordings from the prefrontal cortex (PFC) of behaving rats. Optimized bandwidth estimates provide a characteristic value for the temporal spiking structure of single units and can be modeled as a function of the temporal precision within spiking patterns accounting for the signal-to-noise ratio in simulated data. The distribution of optimized bandwidth estimates of PFC units and their joint distribution with further spike train metrics allows to segregate groups of cells with distinct spiking properties. Additionally, optimized KDEs obtained with UCV-based bandwidths perform reliable or superior compared to non-optimized KDEs when decoding behavioral events during the task. Moreover, when applied to analyze mechanisms of encoding and internal processing during self-paced cognitive tasks, optimized KDEs facilitate across-trial comparisons of firing activity during trials varying in length, enable to identify neuronal ensembles encoding for task-related events and can unfold population dynamics displaying the underlying neural process.

Document type: Dissertation
Supervisor: Durstewitz, Prof. Dr. Daniel
Date of thesis defense: 16 November 2015
Date Deposited: 07 Dec 2015 14:20
Date: 2015
Faculties / Institutes: The Faculty of Bio Sciences > Dean's Office of the Faculty of Bio Sciences
About | FAQ | Contact | Imprint |
OA-LogoDINI certificate 2013Logo der Open-Archives-Initiative