[Koglist] [CEU Cogsci Talks] REMINDER: Invitation to the Doctoral Defense of Dominik Garber at the Department of Cognitive Science, Vienna, May 3, 4 pm CET

Dear All, This is a kind reminder: The Department of Cognitive Science cordially invites you to the public defense of the PhD thesis ABSTRACTION, CONSOLIDATION, AND EXPLICITNESS IN SPATIO-TEMPORAL VISUAL STATISTICAL LEARNING by Dominik Garber FriDAY, May 3, 4 P.M. CET| Room D001 (CEU, Quellenstrasse 51, 1100 Vienna) (Zoom: Meeting: https://ceu-edu.zoom.us/j/91942335617?pwd=RGR4ZjFjaXlmQWxDdzRaUTcyclF1dz09&… Meeting ID: 919 4233 5617 Passcode: 632589 PRIMARY SUPERVISOR: József Fiser (CEU) SECONDARY SUPERVISOR: Máté Lengyel (CEU/Cambridge) Members of the Dissertation Committee: Eva Wittenberg, Chair, CEU Professor <https://www.sheffield.ac.uk/music/people/academic-staff/renee-timmers> Nicolas Turk-Browne<https://psychology.yale.edu/people/nick-turk-browne>ne>, External examiner, Yale University, and Professor Takeo Watanabe<https://www.brown.edu/academics/cognitive-linguistic-psychologi…be>, External examiner Brown University *Anyone not affiliated with CEU wishing to attend in-person in Vienna must RSVP here<https://forms.office.com/e/LK13S1Fzvy> to get access to the lecture hall. ABSTRACT |Visual statistical learning (VSL) describes how humans automatically and implicitly become sensitive to the statistics of visual input in the absence of supervision or reinforcement. Research on VSL usually focuses on learning either temporal or spatial regularities and almost always excludes the influence of prior knowledge. In this dissertation, I present a reconceptualization of VSL as part of a larger human unsupervised learning system operating by combining lower-level spatio-temporal co-occurrence statistics and higher-level top-down biases. I identified three types of higher-level biases affecting statistical learning: (1) pre-existing biases independent of properties of the experiment, (2) biases formed based on the abstraction of learned low-level statistics, and (3) biases based on observed higher-level features of the input. Furthermore, I identified important moderators of this hierarchical learning system: explicit-ness and consolidation of knowledge. Extending the classical spatial VSL paradigm to a transfer learning paradigm, I found that while participants with explicit knowledge could immediately abstract from their acquired representations and generalize to novel input, participants with implicit knowledge showed a structural novelty effect in immediate transfer. This means they were better at learning novel input that was not aligned with what they had learned before. However, after a period of asleep consolidation, participants with implicit knowledge switched their behavior and showed generalization, as the participants with explicit knowledge did before. Using control experiments, I confirmed that this effect is specific to sleep and could not be explained simply by time passing or a time-of-day effect. Furthermore, using matched sample analysis, I demonstrated that differences in the strength of initial learning cannot explain the qualitative differences found between participants with explicit and implicit knowledge. In order to combine the previously disjoint lines of spatial and temporal VSL, I developed a novel spatio-temporal visual statistical learning paradigm. There, spatially defined patterns were unfolding to the observer over time. I demonstrated that implicit learning is possible for spatio-temporal input and provided experimental evidence that the temporal statistics of the input were used for the implicit acquisition of spatial patterns. Furthermore, I showed that when confronting participants with the complexity of spatio-temporal input, top-down, bottom-up interactions naturally emerged, linking this line of research with the VSL transfer learning paradigm described above. I found that both the overall motion direction and the overall arrangement of shapes can bias participants learning and their beliefs about what types of structures are present in the input. Furthermore, by combining the spatio-temporal VSL paradigm with a prediction task, I found that participants with explicit knowledge but not participants with implicit knowledge can use it for prediction, adding to the findings on differences between explicit and implicit representations described above. Overall, this dissertation demonstrates that the narrow limitation and control that enabled the initial success of VSL research need to be carefully and incrementally overcome to understand the role of VSL in the overall human cognitive system. It does so by introducing two new VSL paradigms that enable novel, systematic ways of investigating the human unsupervised learning system. ------------------------------------------------------------------------- Hosted by the Department of Cognitive Science [cid:image001.png@01DA9AF9.F5C4CAA0] ______________________________________________ Subscribe by sending an empty mail to talks-subscribe(a)cogsci.ceu.edu Unsubscribe by sending an empty mail to talks-unsubscribe(a)cogsci.ceu.edu