CEU Electronic Theses and Dissertations, 2025
| Author | Koblinger, Ádám Ferdinánd |
|---|---|
| Title | In Search of Uncertainty: Representations of Uncertainty Across Various Levels of Perception and Cognition |
| Summary | Uncertainty is an inevitable companion of life in our complex world. It is increasingly believed that to act effectively under such circumstances, humans and animals rely on approximately probabilistic computation and an internal model that becomes attuned to environmental regularities and can efficiently complement the incomplete sensory observations with insights distilled from past experiences. However, for efficient behavior, the internal model must maintain veridical information about its own uncertainty. While there is increasing evidence that humans and animals are aware of the uncertainty associated with their decisions, the extent of their uncertainty representations is unclear. Specifically, it is unknown whether they represent uncertainty in a task-dependent manner, solely at the level of decisions, or in a fully Bayesian manner, representing uncertainty just about every aspects of their internal model. In this dissertation, I address this gap by first arguing for the preeminence of fully Bayesian models in terms of their generalization abilities over the alternative candidates. Then, I present three studies that clarify different characteristics of human and animal internal models relevant to assessing the degree to which uncertainty is encoded in biological internal models. Chapter 1 provides a normative justification for the use of fully Bayesian representations, demonstrating their superior data and memory efficiency compared to task-dependent representations. I critically review the literature, highlighting the lack of conclusive evidence on the extent of the brain’s uncertainty representation, and propose experimental paradigms to address this gap, setting the stage for the subsequent chapters. In Chapter 2, I present experimental evidence based on a novel behavioural paradigm that human internal models meet one of the fundamental prerequisites for fully Bayesian models: they simultaneously represent uncertainties about multiple internal variables. Moreover, I found that explicit uncertainty reports about a variable (e. g., orientation) are based on gradually emerging probabilistic perceptual representation of that variable rather than on other, related variables (e.g., contrast) that can serve as proxies for the sensory reliability. In Chapter 3, I use another novel behavioural paradigm and ideal observer analysis to demonstrate that humans automatically employ complex internal models with multiple variables and parameters, even in simple decision-situations where such complexity may not seem necessary. These complex internal models are then updated in a Bayesian manner when changes in task statistics are encountered, providing further support for the fully Bayesian brain hypothesis. In Chapter 4, I propose a novel hybrid experimental paradigm combining neural and behavioral approaches to identify the neural traces of the potential perceptual uncertainty representation that are distinct from the representation of uncertainty directly related to the decision. Next, I demonstrate the practical application of this method to behavioral data from mice performing an orientation estimation task together with neural activity from their primary visual cortex (V1) recorded with calcium imaging. This data provides preliminary evidence that mouse V1 represents perceptual, rather than decision uncertainties, which is encoded in the temporal activity patterns within a trial, rather than in the spatial activity patterns across the population. Together, these results shed a more focused light on the hitherto unexplored extent to which uncertainty is encoded in the brain and provide a consequential support to the proposal that complex brains use complex, approximately probabilistic processes and a broad representation of uncertainties to cope with challenges of their complex and uncertain environment. |
| Supervisor | Fiser, József; Lengyel, Máté |
| Department | Cognitive Science PhD |
| Full text | https://www.etd.ceu.edu/2025/koblinger_adam.pdf |
Visit the CEU Library.
© 2007-2025, Central European University