Research

With: Linda Ventura, Victoria Bosch, Tim Kietzmann

Predicting the next input encodes relational structure amongst the inputs: a RNN path integrates and binds the input tokens to their absolute locations in 2D scenes, in-context.

With: Adrien Doerig, Alexander Kroner, Carmen Amme, Tim Kietzmann

Predicting the next glimpse features (given the saccade) coaxes a network to encode co-occurrence and spatial arrangement in a visual-cortex-aligned scene representation.

With: Lotta Piefke, Adrien Doerig, Tim Kietzmann

In cluttered object tracking with RL, an agent learns an explicit encoding of attentional state (an “attention schema”), most useful when attention cannot be inferred from the stimulus.

With: Adrien Doerig, Tim Kietzmann

The feedforward sweep instantiates a representational arrangement that dovetails with a recurrence-induced “equal movement” prior, enabling corrected classifications.

With: Giacomo Aldegheri, Tim Kietzmann

Recurrent flow carries category-orthogonal feature information (e.g., location) used iteratively to constrain subsequent category inference.

With: Daria Proklova, Daniel Kaiser, Marius Peelen

Animacy organization is not fully driven by visual-feature differences; it also depends on inferred factors like agency quantified behaviorally.

With: Jochem Koopmans, Genevieve Quek, Marius Peelen

In a Sperling-like task, people report either occasionally-present or absent inverted letters as upright to the same extent => expectation-driven illusions may be post-perceptual, not p-conscious.

With: Zejin Lu, Radoslaw Cichy, Tim Kietzmann

Inspired by the Adaptive Initial Degradation hypothesis, ANNs trained with a graded coarse-to-fine image diet produce strong shape-biased, plus distortion/adversarial robust classification behavior.

With: Rowan Sommers, Daniel Anthes, Tim Kietzmann

Inspired by Hummos, we built an image-based Wisconsin Card Sorting Task variant and found behavior suggesting sparks of cognitive flexibility: compositional rule inference in activity space.

With: Daniel Anthes, Peter König, Tim Kietzmann

Readout misalignment due to learning-induced drift is a core continual-learning problem. Constraining drift to the readout null-space helps networks stay both stable and plastic.

With: Johannes Singer, Radoslaw Cichy, Tim Kietzmann

Attentional routing can push task-relevant information through the network as effectively as “direct access,” challenging the necessity of direct access for explaining behavior.

With: Lu-Chun Yeh, Marius Peelen

The relation of multi-object display processing to isolated-object representations is task-dependent: same/different relates to earlier; object search, to later representations in MEG/fMRI.

With: Surya Gayet, Elisa Battistoni, Marius Peelen

Search templates encode identity and size; size is inferred from location in scenes, and is entangled with identity in the template.

With: Genevieve Quek, Marius Peelen

Increased search efficiency amongst co-occurring distractors likely reflects faster/more accurate rejection of a distractor’s partner as a possible target.

With: Marius Peelen

Feature-based attention modulates fMRI representations of body silhouettes presented in task-irrelevant locations in high-level visual cortex.

With: Giacomo Aldegheri, Marcel van Gerven, Marius Peelen

Early bias/gain modulation alleviates later capacity limits; optimized modulations look like tapping a superposition of networks rather than classic feature-similarity gain.

With: Ilze Thoonen, Sjoerd Meijer, Marius Peelen

Scene co-variation biases categorization, but across 4 experiments we found no evidence that task-irrelevant scenes boost sensitivity for detecting co-varying objects.

With: Victoria Bosch, Daniel Anthes, Adrien Doerig, Peter Konig, Tim Kietzmann

fMRI responses to natural scenes condition word generation in a Transformer, enabling flexible readout of semantic properties like object class and numerosity.

With: Varad Choudhari

Reverse dictionary via n-hop reverse search on a definition graph. Matched then-SOTA on ~3k lexicon; didn't scale well to ~80k.

With: Sukanya Patil, Bipin Rajendran

Engineered SNN modules for polynomial transformations on spike trains. Temporal lag hinders SNN-translation of symbolic forward model for velocity–waypoint navigation.