Anticipatory gaze strategies as a principle of active information sampling
Speaker: Lynn Schmittwilken PhD (Perception)
Abstract: We often think about perception as a rather passive process: visual information enters through the eyes and is then analyzed by the brain. However, in natural behavior, the signals that reach the eyes depend strongly on how we move through the world. In this talk, I will explore this perspective using eye movements as an example. I will show how perception is shaped not only by the sensory input, but also by the actions through which this input is gathered.
Eye movements are known to reflect an interplay between task demands and stimulus properties. It remains unclear, however, to what extent they are simple reactions to the visual input or whether they are adjusted in advance to support efficient encoding. To address this question, our work combines behavioral experiments with computational modeling. We study how people move their eyes while searching for visual information under different task demands and levels of visibility. This approach allows us to characterize gaze behavior and to test possible computational principles behind them.
By studying eye movements as part of an active sampling process, this work aims to clarify how action and perception interact. More broadly, it supports the idea that anticipatory gaze strategies are a principle of active information sampling and highlights the role of self-generated motion in visual perception.

Feature-based attention as iterative Bayesian adaptation of latent representations
Speaker: Dr. Johann Bauer (Computational Modelling of Intelligent Systems)
Abstract: Perceptual accuracy often improves with processing time, even for static stimuli that provide no additional sensory evidence. While this speed–accuracy trade-off is well documented behaviorally, its representational and computational basis remains unclear: what changes over time when the input itself is unchanged?
We propose that additional processing time enables iterative attentional re-representation of sensory inputs, allowing a fixed neural architecture to become differentially effective across contexts.
Using a deep neural network as a model of visual inference, we study this idea in the context of image category discrimination as a concrete instance of a general decision problem. Expanding prior work on feature-based attention (e.g., Martinez-Trujillo & Treue, 2004; Lindsay & Miller, 2018), we propose a feedback-driven attentional mechanism trained to reduce entropy over competing hypotheses by modulating intermediate upsteam representations, locally improving the discriminability of representations.
We find that this iterative re-representation yields higher inference accuracy for static inputs, providing a functional and mechanistic account of how processing time improves performance and further presenting a natural mechanism for preparatory feature attention, enabling faster and more accurate inference even at initial input processing, with implications beyond object recognition for hypothesis-driven perception and decision making.

Repeated visuomotor delay adaptation
Speaker: Celine Honekamp M.Sc. (Sensorimotor Control and Learning)
Abstract: When performing a motor task we are often exposed to mismatches between our intended movement behaviour and the outcomes of those actions. In the spatial domain, it was shown that repeated exposures to perturbations lead to behavioural benefits (savings) in the form of a reduced initial error and faster adaptation when being re-exposed. In my work, I instead focus on the temporal domain and adaptation to visuomotor delays. In three VR experiments using either target interception or continuous tracking as the adaptation task, we showed evidence for delay adaptation, negative after-effects and behavioral benefits (savings). Additionally, we showed distinct patterns in terms of the perceived timing shift indicating that different processes in adaptation might be involved for the two tasks.