Neuronale Kodierung und Dekodierung von multiplen und dynamischen Reizen
|Other Titles:||Neuronal Coding and Decoding of Multiple and Dynamic Stimuli||Authors:||Schulzke, Erich Lutz Luciano Franz||Supervisor:||Eurich, Christian||1. Expert:||Eurich, Christian||2. Expert:||Pawelzik, Klaus||Abstract:||
Since the time of Hubel and Wiesel's famous experiments, the investigation of neural response properties with a stimulation paradigm using single static stimulus values has become extremely popular. Theoretical investigations, which were designed according to the experimental paradigm, focussed on the coding accuracy of single stimuli by neural populations, thereby establishing the standard model of population coding. In conjunction with methods based mainlyon Bayesian decoding techniques or estimation-theoretic measures such as Fisher-Information, it has been used to identify coding strategies for optimal encoding of single static stimulus values (localization).However, the world does not consist only of single static stimulus values, resulting in more complex situations to be analyzed by the brain. Therefore, starting from the standardmodel of population coding we explore two extensions regarding the stimulation. The first extension consists in allowing not only a single stimulus value, but arbitrarystimulus distributions at a time. Here we present a framework which allows for modeling situations with multiple stimuli. Model investigations in part I focus on situations with two simultaneously presented stimuli and use Maximum-Likelihood approaches in the compoundstimulus space, which comprises all possible sets of values in the now two-dimensional stimulus space. We introduce a new parameter lambda in the encoding model for two stimuli, which represents a nonlinearity in the interaction between the two average response rates to stimulus values when presented exclusively. In analogy to the analysis for stimulus localization we investigate coding strategies for optimal discrimination between the presence of a single or two stimulus values. We find, that the same strategies hold for both localization and discrimination between a single and two stimulus values. Furthermore, we investigate the influence of an attentional parameter kappa on the encoding accuracy of two stimulus values, where only one is attended.We find that,with increasing kappa, encoding of the attended value becomes more accurate while it becomes worse for the unattended value. This effect can be modulated by lambda, hereby suggesting a possible role for lambda in allowing to focus on the behaviorally relevant stimulus. We also apply our framework to the analysis of the motion repulsion effect, a psychophysical phenomenon, where observers systematically misinterpret the difference between the directions of two moving superimposed random dot clouds. We model the conditions during the psychophysical experiments and assume false beliefs, a lack of knowledge of the decoding population about the parameters during encoding. Under such conditions, the model is able to reproduce repulsion effects in a similar quantitative range as in the experiments. Finally, regarding the estimation of a bimodal stimulus distribution, we find that small tuning curves improve the decoding accuracy, as found for single stimulus values. The model is able to make testable predictions for different situations and can be readily applied to more situations involving multiple stimuli to explain e.g. the metamers found by S. Treue and colleagues. It may prove to be a useful tool in unraveling the computational principles underlying neural computation.The second extension to the standard model is concerned with stimulus dynamics. In part II, the electrophysiologically recorded responses of neurons from visual areal MT in the macaque to dynamic stimuli with different temporal statistics are analyzed. We test if neural responsecharacteristics depend upon stimulus statistics, adapting on a timescale of seconds. The analysis employs a virtual population of LNP-neurons with static response characteristics, where the latter is stimulated by the same stimuli presented to the macaque, and finds that differences in response characteristics can be fully explained by feed-forward stimulus influence. Reconstruction accuracy of the populations depends heavily on stimulus statistics,where the width of the integration time proves to be the most crucial parameter for the encoding of dynamic stimuli.
|Keywords:||Neuronal Coding, Neuronal Decoding, Bayesian Decoding, Area MT, Multiple Stimuli, Dynamic Stimuli, Stimulus Reconstruction, Macaque||Issue Date:||26-May-2006||URN:||urn:nbn:de:gbv:46-diss000103601||Institution:||Universität Bremen||Faculty:||FB1 Physik/Elektrotechnik|
|Appears in Collections:||Dissertationen|
checked on Oct 26, 2020
checked on Oct 26, 2020
Items in Media are protected by copyright, with all rights reserved, unless otherwise indicated.