Gergő Orbán

This paper develops a hierarchical deep generative model that explains top-down influences in early visual cortex by predicting neural activity in V1 and V2.

This study identifies a neural subspace in primate frontal cortex that tracks foraging-like computations, showing both foraging and reinforcement learning signatures.

The paper shows attention enhances decoding of natural scenes in V1 by aligning with low-dimensional subspaces of natural stimuli, but not synthetic ones.

The paper shows that V1 neurons represent both visual stimuli and task context in overlapping but orthogonal population activity patterns during a cross-modal decision task.

The paper shows that anterior cingulate cortex encodes stimuli in low-dimensional subspaces during decision making, with context-gated suppression of irrelevant stimuli.

The paper investigates how uncertainty is represented in hippocampal theta sequences during navigation and finds that uncertainty is encoded sequentially through sampling motion trajectories rather than encoding probability distribution parameters.

The paper proposes a semantic compression framework to explain memory distortions by modeling memory as lossy compression using generative models, validated across three datasets.

The paper infers individualized internal models from response times in a visuomotor learning task, revealing structured inductive bias varying across individuals.

The paper shows that stimulus-specific spike count correlations in V1 are shaped by feedback interactions and high-order stimulus structure, independent of response means.