Projects

I work in the lab of Christian Ruff at the Zurich Center for Neuroeconomics, where most of my current projects are based. On this page, you will find some information on ongoing and former projects:

Ongoing

• How perception shapes economic choice
• From perception to value
• Attention and visual receptive fields

Past

• Ocular dominance columns, retinotopy and cortical depth
• The human subcortex at ultra-high field

How perception shapes economic choice

Collaborators: Miguel Barretto-García, Marius Moisa, Maike F. Renkert, Arthur Prat-Carrabin, Saurabh Bedi, Marcus Grueschow, Rafael Polanía, Michael Woodford, Franciszek Hennel, Samuel J. Gershman, Christian C. Ruff

This is the core line of my work in the lab of Christian Ruff. When choosing between a safer and a riskier option, people tend to prefer the safer one, even when the riskier option pays more on average. Classical economic theories like expected utility and prospect theory explain this through valuation: people do not like risk.

We explore a complementary hypothesis: that risk aversion is (also) rooted in perception. In many domains, such as numerosity, loudness, weight and brightness, people tend to underestimate large magnitudes. So why would this be different for money? This idea is formalized in a theoretical framework developed by Khaw, Li and Woodford.

(a) Numerical receptive field models show that patches of parietal cortex prefer specific numbers, laid out as maps. (b) These populations are tuned on a logarithmic scale. (c) A noisy log code plus prior knowledge makes large payoffs look smaller than they are, more so when the code is noisier. (d) As a result, people need a larger premium before they take a gamble: risk aversion without any dislike of risk. Panels b–d are model illustrations.

We use these numerical population receptive field models, which build on seminal work by Ben Harvey and Serge Dumoulin, to link choices to the way parietal cortex represents magnitude. So far, we have shown that:

• Noisier representations go with more risk aversion. People with noisier numerical representations in parietal cortex are more risk-averse (Barretto-García*, de Hollander* et al., 2023, Nature Human Behaviour).
• Risk attitudes change from moment to moment. Payoffs held in working memory are noisier and pulled towards prior beliefs, which can shift people between risk aversion and risk seeking (de Hollander et al., 2026, Nature Communications).
• The link is causal. Combining TMS with fMRI, we found that disrupting parietal cortex weakens numerical tuning and increases risk seeking (de Hollander*, Moisa* & Ruff, 2025, preprint).
• Stress acts on perception. Acute stress reduces risk aversion by changing how payoff magnitudes are perceived, rather than how they are valued (Renkert*, de Hollander* et al., 2025, preprint).
• Number codes adapt to context. Numerical receptive fields shift and rescale with the range of relevant numbers, a form of efficient coding on the fly (Prat-Carrabin*, de Hollander* et al., 2026, Nature Communications). You can explore them in our 3D brain viewers.
• Number codes differ in dyscalculia. Dyscalculia reflects less precise numerical representations in parietal cortex and an altered allocation of this cortex to functional networks (Renkert et al., 2026, preprint).

We are now extending this work to new questions, such as how parietal cortex handles several numbers at once.

From perception to value

Collaborators: Saurabh Bedi, Raphaël Le Bouc, Maximilian V. Harl, Marcus Grueschow, Shira M. Lupkin, Vincent B. McGinty, Rafael Polanía, Christian C. Ruff

If limited neural resources shape how we represent numbers, the same should hold for other quantities, and for later stages of the decision process. In work led by Saurabh Bedi, we asked whether perception and valuation each have their own efficient code.

By varying how stimulus orientation maps onto value, we could tease apart efficient codes in perception and valuation (Bedi et al., 2026).

• Perception and value have separable codes. In a preregistered study, behavior was best explained by efficient coding at both stages, each tracking its own prior. Changing the value distribution even reversed the classic repulsion biases of orientation perception. Perceptual codes reflected stable, long-term structure, while value codes updated rapidly with context (Bedi et al., 2026, preprint).
• Probability distortions come from noise near boundaries. People overweight small and underweight large probabilities because cognitive noise is repelled by the natural boundaries of probability, 0 and 1. Three preregistered experiments confirmed the predictions of this account (Bedi, de Hollander & Ruff, 2025, preprint).
• Value itself has a population code. The OFC/vmPFC implements a probabilistic population code for value, and its decoded uncertainty predicts preference instability and confidence (Le Bouc et al., 2026, preprint).

We are now studying how these codes change as people learn what stimuli are worth.

Attention and visual receptive fields

Collaborators: Marco Aqil, Martin Szinte, Serge Dumoulin, Tomas Knapen, Dock Duncan, Jan Theeuwes

Humans can actively process much less information than they have access to. Imagine crossing a busy street: you probably observe different aspects of the traffic one after the other before you commit to crossing. This moving ‘‘spotlight’’ is a form of selective attention. But how does it change the way the brain represents the world?

By inverting population receptive field models, we can calculate the position of a stimulus given corresponding neural signals in the visual system. Moreover, we can also quantify the certainty we have about this position and thereby, possibly, the reliability of the underlying neural signals.

We first fit population receptive field models to see what part of the brain represents what part of the visual field. Then, we invert those models to see ‘‘what the brain is representing’’ during a task, and with how much uncertainty. These methods are implemented in our braincoder package.

• Vision stays retinotopic. Receptive fields shift with gaze, so space is represented in a retinotopic reference frame throughout human visual cortex (Szinte*, de Hollander*, Aqil* et al., 2026, eLife reviewed preprint).
• Psilocybin changes visual context effects. Psilocybin alters contextual perception in the Ebbinghaus illusion, as well as contextual modulation in cortical responses (Aqil, de Hollander et al., 2025, Nature Communications).

Together with Dock Duncan and Jan Theeuwes at VU Amsterdam, we are now studying how reward and past experience reshape visual receptive fields.

Ocular dominance columns, retinotopy and cortical depth

Collaborators: Tomas Knapen

Using the increased SNR of 7T MRI combined with surface coils (built with smaller coil elements that are placed very close to the skull), we can obtain functional MRI at an unprecedented resolution of 0.7 mm. This allowed us to measure individual maps of visual cortex that describe how the outside world is represented. Specifically, we visualized both *ocular dominance columns (ODCs) and retinotopic maps. So for each patch of primary visual cortex, we could visualize (1) to which part of the visual field and (2) to which single eye that patch of cortex was most sensitive.

In this experiment, we mapped out the sensitivity of neurons in primary visual cortex (V1) to different parts of the visual field, as well as different eye using fMRI. (A) our field of view only included the occipital pole and thereby the foveal parts of V1. The data was carefully registered to a 0.64mm T1-weighted anatomical image. (B) We performed a standard PRF mapping paradigm and PRF model to quantify to which parts of the visual field different patches of V1 were most sensitive. This allowed us to relate columnar patterns to retinotopy, as well as define the border between V1 and V2 by the characteristic phase reversals in retinotopy. (C) By presenting flickering checkerboards separately to the two eyes, contained by a salient frame that helped eye convergence, we could map out ocular dominance columns: little ~1mm-wide stripes of corte that are particularly sensitive to the left or right eye.

You can explore the ocular dominance column maps of three participants on their own cortical surfaces in these interactive brain maps.

Moreover, due to the 0.7mm resolution, we could also dissociate fMRI activity across different laminae. Specifically, we could show that whereas differential input to the eyes mainly affected the fMRI signal at deep and middle depth, corresponding to infragranular and granular layer, an attentional manipulation almost exclusively impacted signal at superficial depth, corresponding to supragranular layers. Crucially, these findings nicely mirror findings in electrophysiological recordings in rodents and non-human primates, as well as neuroanatomical tracing studies that show that feedback projections largely terminate in supragranular layers, whereas feedforward and sensory input largely terminates in granular and infragranular layers.

In sum, this project was a nice illustration of the potential of 7T fMRI to move beyond golfball-sized “blobs” in an MNI standard space, and move towards individualized neuroimaging that focuses on individual representations of the outside world.

The human subcortex at ultra-high field

Collaborators: Birte U. Forstmann, Max C. Keuken, Anneke Alkemade, Steven Miletić

During my PhD with Birte Forstmann at the University of Amsterdam, I studied small subcortical structures that are hard to image, in particular the subthalamic nucleus (STN), using 7 Tesla MRI, histology and computational models like the drift diffusion model. Much of the human subcortex is still poorly understood (Forstmann et al., 2017, Nature Reviews Neuroscience).

• Iron changes gradually within the STN. Iron increases gradually towards its medial-inferior tip, as we showed with quantitative susceptibility mapping in vivo and post mortem (de Hollander*, Keuken* et al., 2014, Human Brain Mapping).
• Reported STN activity is often misplaced. Reported ‘‘STN’’ activations are often a mix of the STN, the substantia nigra and their surroundings (de Hollander*, Keuken* et al., 2015, PLOS ONE), and basal ganglia fMRI needs dedicated protocols at 7T (de Hollander et al., 2017, Human Brain Mapping).
• The STN shows gradients, not discrete parts. Histological markers show gradients rather than discrete subdivisions (Alkemade*, de Hollander*, Miletić* et al., 2019, Brain Structure and Function).

You can read the full story in my PhD thesis.