On the relationship between neuronal codes and mental models

DSpace Repositorium (Manakin basiert)

Zur Kurzanzeige

dc.contributor.advisor Mallot, Hanspeter Anton (Prof. Dr.)
dc.contributor.author Ecke, Gerrit
dc.date.accessioned 2022-02-22T15:36:35Z
dc.date.available 2022-02-22T15:36:35Z
dc.date.issued 2022-02-22
dc.identifier.uri http://hdl.handle.net/10900/124763
dc.identifier.uri http://nbn-resolving.de/urn:nbn:de:bsz:21-dspace-1247632 de_DE
dc.identifier.uri http://dx.doi.org/10.15496/publikation-66126
dc.description.abstract Das übergeordnete Ziel meiner Arbeit an dieser Dissertation war ein besseres Verständnis des Zusammenhangs von mentalen Modellen und den zugrundeliegenden Prinzipien, die zur Selbstorganisation neuronaler Verschaltung führen. Die Dissertation besteht aus vier individuellen Publikationen, die dieses Ziel aus unterschiedlichen Perspektiven angehen. Während die Selbstorganisation von Sparse-Coding-Repräsentationen in neuronalem Substrat bereits ausgiebig untersucht worden ist, sind viele Forschungsfragen dazu, wie Sparse-Coding für höhere, kognitive Prozesse genutzt werden könnte noch offen. Die ersten zwei Studien, die in Kapitel 2 und Kapitel 3 enthalten sind, behandeln die Frage, inwieweit Repräsentationen, die mit Sparse-Coding entstehen, mentalen Modellen entsprechen. Wir haben folgende Selektivitäten in Sparse-Coding-Repräsentationen identifiziert: mit Stereo-Bildern als Eingangsdaten war die Repräsentation selektiv für die Disparitäten von Bildstrukturen, welche für das Abschätzen der Entfernung der Strukturen zum Beobachter genutzt werden können. Außerdem war die Repräsentation selektiv für die die vorherrschende Orientierung in Texturen, was für das Abschätzen der Neigung von Oberflächen genutzt werden kann. Mit optischem Fluss von Eigenbewegung als Eingangsdaten war die Repräsentation selektiv für die Richtung der Eigenbewegung in den sechs Freiheitsgraden. Wegen des direkten Zusammenhangs der Selektivitäten mit physikalischen Eigenschaften können Repräsentationen, die mit Sparse-Coding entstehen, als frühe sensorische Modelle der Umgebung dienen. Die kognitiven Prozesse hinter räumlichem Wissen ruhen auf mentalen Modellen, welche die Umgebung representieren. Wir haben in der dritten Studie, welche in Kapitel 4 enthalten ist, ein topologisches Modell zur Navigation präsentiert, Es beschreibt einen dualen Populations-Code, bei dem der erste Populations-Code Orte anhand von Orts-Feldern (Place-Fields) kodiert und der zweite Populations-Code Bewegungs-Instruktionen, basierend auf der Verknüpfung von Orts-Feldern, kodiert. Der Fokus lag nicht auf der Implementation in biologischem Substrat oder auf einer exakten Modellierung physiologischer Ergebnisse. Das Modell ist eine biologisch plausible, einfache Methode zur Navigation, welche sich an einen Zwischenschritt emergenter Navigations-Fähigkeiten in einer evolutiven Navigations-Hierarchie annähert. Unser automatisierter Test der Sehleistungen von Mäusen, welcher in Kapitel 5 beschrieben wird, ist ein Beispiel von Verhaltens-Tests im Wahrnehmungs-Handlungs-Zyklus (Perception-Action-Cycle). Das Ziel dieser Studie war die Quantifizierung des optokinetischen Reflexes. Wegen des reichhaltigen Verhaltensrepertoires von Mäusen sind für die Quantifizierung viele umfangreiche Analyseschritte erforderlich. Tiere und Menschen sind verkörperte (embodied) lebende Systeme und daher aus stark miteinander verwobenen Modulen oder Entitäten zusammengesetzt, welche außerdem auch mit der Umgebung verwoben sind. Um lebende Systeme als Ganzes zu studieren ist es notwendig Hypothesen, zum Beispiel zur Natur mentaler Modelle, im Wahrnehmungs-Handlungs-Zyklus zu testen. Zusammengefasst erweitern die Studien dieser Dissertation unser Verständnis des Charakters früher sensorischer Repräsentationen als mentale Modelle, sowie unser Verständnis höherer, mentalen Modellen für die räumliche Navigation. Darüber hinaus enthält es ein Beispiel für das Evaluieren von Hypothesn im Wahr\-neh\-mungs-Handlungs-Zyklus. de_DE
dc.description.abstract The superordinate aim of my work towards this thesis was a better understanding of the relationship between mental models and the underlying principles that lead to the self-organization of neuronal circuitry. The thesis consists of four individual publications, which approach this goal from differing perspectives. While the formation of sparse coding representations in neuronal substrate has been investigated extensively, many research questions on how sparse coding may be exploited for higher cognitive processing are still open. The first two studies, included as chapter 2 and chapter 3, asked to what extend representations obtained with sparse coding match mental models. We identified the following selectivities in sparse coding representations: with stereo images as input, the representation was selective for the disparity of image structures, which can be used to infer the distance of structures to the observer. Furthermore, it was selective to the predominant orientation in textures, which can be used to infer the orientation of surfaces. With optic flow from egomotion as input, the representation was selective to the direction of egomotion in 6 degrees of freedom. Due to the direct relation between selectivity and physical properties, these representations, obtained with sparse coding, can serve as early sensory models of the environment. The cognitive processes behind spatial knowledge rest on mental models that represent the environment. We presented a topological model for wayfinding in the third study, included as chapter 4. It describes a dual population code, where the first population code encodes places by means of place fields, and the second population code encodes motion instructions based on links between place fields. We did not focus on an implementation in biological substrate or on an exact fit to physiological findings. The model is a biologically plausible, parsimonious method for wayfinding, which may be close to an intermediate step of emergent skills in an evolutionary navigational hierarchy. Our automated testing for visual performance in mice, included in chapter 5, is an example of behavioral testing in the perception-action cycle. The goal of this study was to quantify the optokinetic reflex. Due to the rich behavioral repertoire of mice, quantification required many elaborate steps of computational analyses. Animals and humans are embodied living systems, and therefore composed of strongly enmeshed modules or entities, which are also enmeshed with the environment. In order to study living systems as a whole, it is necessary to test hypothesis, for example on the nature of mental models, in the perception-action cycle. In summary, the studies included in this thesis extend our view on the character of early sensory representations as mental models, as well as on high-level mental models for spatial navigation. Additionally it contains an example for the evaluation of hypotheses in the perception-action cycle. en
dc.language.iso en de_DE
dc.publisher Universität Tübingen de_DE
dc.rights ubt-podok de_DE
dc.rights.uri http://tobias-lib.uni-tuebingen.de/doku/lic_mit_pod.php?la=de de_DE
dc.rights.uri http://tobias-lib.uni-tuebingen.de/doku/lic_mit_pod.php?la=en en
dc.subject.ddc 004 de_DE
dc.subject.ddc 500 de_DE
dc.subject.ddc 570 de_DE
dc.subject.other sparse coding de_DE
dc.subject.other Independent component analysis de_DE
dc.subject.other independent component analysis en
dc.subject.other computational neuroscience en
dc.subject.other sparse coding en
dc.subject.other cognitive neuroscience en
dc.subject.other navigation en
dc.subject.other biomimetic en
dc.title On the relationship between neuronal codes and mental models de_DE
dc.type PhDThesis de_DE
dcterms.dateAccepted 2021-06-24
utue.publikation.fachbereich Biologie de_DE
utue.publikation.fakultaet 7 Mathematisch-Naturwissenschaftliche Fakultät de_DE
utue.publikation.noppn yes de_DE

Dateien:

Das Dokument erscheint in:

Zur Kurzanzeige