|
Project C1: Neural systems underlying curiosity-driven sampling of perceptual information PRs: Caspar Schwiedrzik, Fabian Sinz Project C1 will investigate the neural systems that underly the motor actions driving the curious-sampling of perceptual information in the environment. The PhD researcher will, together with the PIs, devise a set of studies to explore how curiosity functions as a mechanism for active learning; the aim of the project is to gather neurophysiological evidence that will help us resolve question (C) - How are we curious? To achieve this aim, the PhD researcher will have video-based eye-tracking, Electroretinograms, EEG and fMRI methods at their disposal.
|
 |
|
Project C2: Curious visual representation learning in children and machines PRs: Alexander Ecker, Nivedita Mani Project C2 will investigate active vision during development under realistic and naturalistic conditions. The overarching goal of the project is to understand what guides infants and children in their selection of information – for example which features of an object children attend to when learning about object categories. To tackle this question, the PhD researcher will train artificial learning systems in active vision tasks and compare the learned information sampling strategies of these systems to those of children of different age groups.
|
 |
|
Project C3: The mechanistics of local-learning rules for curiosity in neural network models PRs: Viola Priesemann Project C3 will explore how and under which conditions neural networks should change their learning rules and strategies to optimally implement curious learning. The PhD researcher will work together with the PIs and the interdisciplinary team, and will contribute to shaping studies in which we (i) optimize the neural implementation of efficient, curious learning based on learning rules derived from information-theoretic principles, (ii) explore how learning rules are adapted, depending on whether an agent is exploring a novel or a known environment, and/or (iii) investigate how this translates into differences in the processing of novel and familiar information.
|
 |
|
Project C4: Modelling curiosity constrained by planning PRs: Florentin Wörgötter Project C4 will investigate how planning towards reaching a certain goal influences curiosity-driven exploration. The PhD researcher will theoretically and experimentally explore (i) how overt and covert biases associated with different potential path options influence exploration in a planning problem, and (ii) how prior experience in one planning problem may influence exploration behavior in a second related problem. The PhD researcher will initially focus on how agents approach simple planning problems, however, there is the potential to work up to studies that employ more complex planning structures (e.g., structures with multiple branches that differ in terms of cost and complexity).
|
 |
|
Project C5: Information theoretic models of curiosity in hierarchichal models of the world PRs: Fabian Sinz, Nivedita Mani Project C5 will investigate whether the information-theoretic ideas of novelty and uncertainty in internal hierarchical generative models of the world can explain what humans are curious about. The PhD researcher will design a series of computer-game based experiments to test the premise that the elements of a scene that are most informative about (our internal model of) the state of the world, are maximally curiosity grabbing. In light of reported developmental differences in children’s exploratory behavior, we plan to test this premise in young children between the ages of four and seven.
|
 |