Contributions

Preprints

2026

1. 

   Faces and bodies are increasingly integrated along the visual hierarchy in humans and deep neural networks

   van Dyck, L. E., and Dobs, K.

   Abs DOI

   Human visual cortex contains regions specialized for faces and bodies, yet we perceive people as a whole. Why does the brain appear to segregate faces and bodies, and how are they integrated to support person perception? Here, we test whether deep neural network models optimized for visual recognition develop segregated or integrated face and body processing, and how this compares to fMRI activity in visual cortex during natural image viewing. We find that models contain face- and body-selective units but also mixed-selective units that are tuned to both faces and bodies. While face- and body-selective units explain unique variance in their corresponding cortical regions, mixed-selective units best explain activity across regions, and shared variance increases from posterior to anterior cortex. Together, our findings suggest that face and body processing reflects a balance of segregation and integration along the visual hierarchy in humans and models, supporting specialized yet flexible person perception.

Papers

2026

1. 

   Multidimensional feature tuning in category-selective areas of human visual cortex

   van Dyck, L. E., Hebart, M. N., and Dobs, K.

   The Journal of Neuroscience

   Abs DOI

   Two prominent accounts describe the functional organization of human high-level visual cortex. A categorical view emphasizes category-selective areas, while a dimensional view highlights continuous feature maps spanning these areas. Here, we asked whether these two views reflect complementary expressions of the same underlying organization. Using a data-driven decomposition of fMRI responses from human participants (female and male) in face-, body-, and scene-selective areas, we identified spatially overlapping activity patterns that were shared across individuals. Each area encoded multiple interpretable dimensions capturing both finer within-category and coarser between-category distinctions, even in the most category-selective voxels. These dimensions formed distinct clusters within category-selective areas but extended as distributed maps across visual cortex. Together, these findings reveal an underlying organization that links category-selective areas to continuous feature maps, thereby reconciling categorical and dimensional accounts of high-level visual cortex.

2025

1. 

   Category selectivity as a window into behavioral relevance

   van Dyck, L. E., and Dobs, K.

   Cognitive Neuroscience

   Abs DOI

   Ritchie et al. (this issue) argue that a deeper understanding of occipitotemporal cortex (OTC) requires shifting emphasis from category selectivity to behavioral relevance. They suggest that focusing on categories such as faces, bodies, or scenes is too narrow and overlooks how OTC supports flexible, goal-directed behavior. We agree that linking neural representations to behavior is essential but caution against treating category selectivity and behavioral relevance as opposing views. Category selectivity provides valuable insight into how cortical representations are organized to support behavior, and recent advances in computational modeling, particularly with deep neural networks, offer a powerful framework for probing this relationship.

2024

1. 

   Artificial intelligence meets body sense: Task-driven neural networks reveal computational principles of the proprioceptive pathway

   van Dyck, L. E., Bremmer, F., and Dobs, K.

   Signal Transduction and Targeted Therapy

   Abs DOI

   In a recent study published in Cell, Marin Vargas and Bisi et al. present an innovative approach to unravel the computational principles underlying proprioceptive processing in non-human primates. Their findings showcase the utility of task-driven modeling in advancing neuroscience and offer translational potential by providing seminal insights into the goals and mechanisms by which the brain encodes body position and movements.

2023

1. 

   Modeling biological face recognition with deep convolutional neural networks

   van Dyck, L. E., and Gruber, W. R.

   Journal of Cognitive Neuroscience

   Abs DOI

   Deep convolutional neural networks (DCNNs) have become the state-of-the-art computational models of biological object recognition. Their remarkable success has helped vision science break new ground, and recent efforts have started to transfer this achievement to research on biological face recognition. In this regard, face detection can be investigated by comparing face-selective biological neurons and brain areas to artificial neurons and model layers. Similarly, face identification can be examined by comparing in vivo and in silico multidimensional “face spaces.” In this review, we summarize the first studies that use DCNNs to model biological face recognition. On the basis of a broad spectrum of behavioral and computational evidence, we conclude that DCNNs are useful models that closely resemble the general hierarchical organization of face recognition in the ventral visual pathway and the core face network. In two exemplary spotlights, we emphasize the unique scientific contributions of these models. First, studies on face detection in DCNNs indicate that elementary face selectivity emerges automatically through feedforward processing even in the absence of visual experience. Second, studies on face identification in DCNNs suggest that identity-specific experience and generative mechanisms facilitate this particular challenge. Taken together, as this novel modeling approach enables close control of predisposition (i.e., architecture) and experience (i.e., training data), it may be suited to inform long-standing debates on the substrates of biological face recognition.

2022

1. 

   Guiding visual attention in deep convolutional neural networks based on human eye movements

   van Dyck, L. E., Denzler, S. J., and Gruber, W. R.

   Frontiers in Neuroscience

   Abs DOI

   Deep Convolutional Neural Networks (DCNNs) were originally inspired by principles of biological vision, have evolved into best current computational models of object recognition, and consequently indicate strong architectural and functional parallelism with the ventral visual pathway throughout comparisons with neuroimaging and neural time series data. As recent advances in deep learning seem to decrease this similarity, computational neuroscience is challenged to reverse-engineer the biological plausibility to obtain useful models. While previous studies have shown that biologically inspired architectures are able to amplify the human-likeness of the models, in this study, we investigate a purely data-driven approach. We use human eye tracking data to directly modify training examples and thereby guide the models’ visual attention during object recognition in natural images either toward or away from the focus of human fixations. We compare and validate different manipulation types (i.e., standard, human-like, and non-human-like attention) through GradCAM saliency maps against human participant eye tracking data. Our results demonstrate that the proposed guided focus manipulation works as intended in the negative direction and non-human-like models focus on significantly dissimilar image parts compared to humans. The observed effects were highly category-specific, enhanced by animacy and face presence, developed only after feedforward processing was completed, and indicated a strong influence on face detection. With this approach, however, no significantly increased human-likeness was found. Possible applications of overt visual attention in DCNNs and further implications for theories of face detection are discussed.

2021

1. 

   Comparing object recognition in humans and deep convolutional neural networks—an eye tracking study

   van Dyck, L. E., Kwitt, R., Denzler, S. J., and Gruber, W. R.

   Frontiers in Neuroscience

   Abs DOI

   Deep convolutional neural networks (DCNNs) and the ventral visual pathway share vast architectural and functional similarities in visual challenges such as object recognition. Recent insights have demonstrated that both hierarchical cascades can be compared in terms of both exerted behavior and underlying activation. However, these approaches ignore key differences in spatial priorities of information processing. In this proof-of-concept study, we demonstrate a comparison of human observers (N = 45) and three feedforward DCNNs through eye tracking and saliency maps. The results reveal fundamentally different resolutions in both visualization methods that need to be considered for an insightful comparison. Moreover, we provide evidence that a DCNN with biologically plausible receptive field sizes called vNet reveals higher agreement with human viewing behavior as contrasted with a standard ResNet architecture. We find that image-specific factors such as category, animacy, arousal, and valence have a direct link to the agreement of spatial object recognition priorities in humans and DCNNs, while other measures such as difficulty and general image properties do not. With this approach, we try to open up new perspectives at the intersection of biological and computer vision research.

2020

1. 

   Predicting engagement in electronic surveillance in romantic relationships

   Hertlein, K. M., and van Dyck, L. E.

   Cyberpsychology, Behavior, and Social Networking

   DOI

Conferences

2026

1. Different features explain category selectivity in human visual cortex and deep neural networks

   van Dyck, L. E., and Dobs, K.

   Cognitive Computational Neuroscience (CCN)
   New York, NY

   Abs DOI

   Category-selective responses to faces, bodies, and scenes are well established in human visual cortex and computational models. However, most studies test only a limited number of categories, leaving it unclear whether selectivity is robust under broader sampling and whether similar features shape selectivity in cortex and models. Here, we measured selectivity for hundreds of categories in ventral temporal cortex (VTC) and deep neural networks (DNNs) and tested which behaviorally relevant dimensions best explain the degree of selectivity across categories. We found that under broad category sampling, selectivity remains robust but varies continuously. In VTC, the most selective categories aligned with classic domains (faces, bodies, scenes), and variability in selectivity was primarily explained by semantic dimensions. In DNNs, by contrast, the most selective categories were visually distinctive animals and objects, and variability in selectivity was explained by mixed visual-semantic dimensions. These findings suggest that selectivity remains robust across a broad category space but is largely shaped by different features in human visual cortex and current computational models.

2. Feature-based principles of category selectivity in human visual cortex and deep neural networks

   van Dyck, L. E., and Dobs, K.

   Vision Sciences Society Meeting (VSS)
   St. Pete Beach, FL
3. Model-guided stimulus curation for comparing artificial and biological vision

   Flachot, A., van Dyck, L. E., and Dobs, K.

   Workshop on CONCEPTS, ACTIONS, and OBJECTS (CAOs)
   Rovereto
4. Face and body representations converge along the visual hierarchy in models and cortex

   van Dyck, L. E., and Dobs, K.

   Workshop on Humans, Deep Networks & Face and Body Recognition
   London

2025

1. Deep neural networks provide insights into distinct and shared selectivity for faces and bodies in human visual cortex

   van Dyck, L. E., and Dobs, K.

   Cognitive Computational Neuroscience (CCN)
   Amsterdam

   Abs DOI

   Faces and bodies are key social stimuli pro-cessed by distinct functional networks in the hu-man visual cortex. However, emerging evidence points to systematic overlap between these net-works, raising the question: How does the brain integrate faces and bodies into whole-person rep-resentations? Competing hypotheses propose fully segregated pathways, varying levels of integration, or a sin-gle multiplexed system. Here, we test these hy-potheses using a deep convolutional neural net-work trained on object recognition. A functional localizer identified face- and body-selective units, as well as mixed-selective units responding to both categories. Decoding analyses revealed that face- and body-selective units specialize in their domains, while mixed-selective units encode de-tailed information from both, supporting a role in face-body integration. Finally, using fMRI encod-ing models, we found that these units uniquely explain neural responses in distinct and overlap-ping face- and body-selective areas in the visual cortex. Our findings suggest that face and body networks balance segregation and integration, supporting both detailed recognition and whole-person perception.

2. Deep neural networks provide insights into distinct and shared selectivity for faces and bodies in human visual cortex

   van Dyck, L. E., and Dobs, K.

   Workshop on CONCEPTS, ACTIONS, and OBJECTS (CAOs)
   Rovereto

2024

1. Core neural dimensions of functionally selective areas in the human visual cortex

   van Dyck, L. E., Hebart, M. N., and Dobs, K.

   Cognitive Computational Neuroscience (CCN)
   Boston, MA

   Abs DOI

   Prior research has extensively documented functional selectivity for categories within visual cortical areas, primarily by contrasting neural responses to images from various categories. However, such categorical approaches are less suitable to capture the diversity of neural representations within these areas. Do category-selective areas encode holistic categories, or are they instead tuned to multifaceted features? To address this question, we employed non-negative matrix factorization (NMF) for analyzing human fMRI responses to natural images in face-, body-, and scene-selective areas, which uncovered a consistent set of interpretable neural dimensions across participants. These dimensions not only aligned with the areas’ respective category preferences, but also revealed finer within-category distinctions, indicating selective tuning to diverse visual input features. Mapping these dimensions onto the cortical surface showed both clustered and distributed topographies, which accounted for overlaps between areas. Our results suggest that category-selective areas show multifaceted feature tuning, challenging traditional views and highlighting the complex interplay of neural dimensions in encoding visual information.

2. Neural representational dimensions capture the nested functional organization of the human visual cortex

   van Dyck, L. E., Hebart, M. N., and Dobs, K.

   SFB Workshop Categorization in Perception and Action: Minds, Models, Mechanisms
   Marburg
3. Core neural dimensions of functionally selective areas in the human visual cortex

   van Dyck, L. E., Dobs, K., and Hebart, M. N.

   Workshop on CONCEPTS, ACTIONS, and OBJECTS (CAOs)
   Rovereto

2022

1. Analyzing and increasing the similarity of humans and deep convolutional neural networks in object recognition

   van Dyck, L. E., Denzler, S. J., and Gruber, W. R.

   European Conference on Visual Perception (ECVP)
   Nijmegen
2. Spatial similarities between human eye movements and deep convolutional neural network saliency maps across time

   van Dyck, L. E., Denzler, S. J., Schöllkopf, C. P., and Gruber, W. R.

   Salzburg Mind-Brain Annual Meeting (SAMBA)
   Salzburg

Theses

2023

1. 

   Unraveling top-down and bottom-up processes in theory of mind with layer fMRI

   van Dyck, L. E.

   University of Salzburg, Department of Psychology

   Abs PDF

   Theory of Mind (ToM), the ability to reason about the mental states of others, is a crucial social skill of humans. Neuroimaging studies have found that the Temporo-Parietal Junction (TPJ) and an entire network of additional brain regions are specifically activated during mental state reasoning. Two common tasks that are used to investigate ToM are the False Belief (FB) task and the Social Animations (SA) task, which are thought to activate the posterior part (pTPJ) and the anterior part (aTPJ) of TPJ, respectively. While previous research has suggested potential explanations for this functional specialization, the exact mechanisms are not yet fully understood. In this study, high-resolution layer fMRI was used to examine neural activity in cortical layers of pTPJ and aTPJ during the FB and SA tasks. Firstly, the results corroborate the importance of the ToM network and especially TPJ in mental state reasoning. Secondly, a significant interaction between task and region was revealed, which underlines the expected functional specialization of TPJ clusters. Thirdly, the layer profiles of the two tasks indicated feedback-like activity, but when separated by region, pTPJ showed feedback-like activity for the FB task, while aTPJ displayed feedforward-like activity for the SA task. This pattern was further confirmed by a hierarchical cluster analysis. Overall, these findings suggest that the functional specialization, which is even reflected at the level of cortical layers, may enable TPJ to switch between detecting social cues externally and contemplating about them internally.

2020

1. 

   Seeing eye-to-eye? A comparison of object recognition performance in humans and deep convolutional neural networks under image manipulation

   van Dyck, L. E.

   University of Salzburg, Department of Psychology

   Abs DOI

   For a considerable time, deep convolutional neural networks (DCNNs) have reached human benchmark performance in object recognition. On that account, computational neuroscience and the field of machine learning have started to attribute numerous similarities and differences to artificial and biological vision. This study aims towards a behavioral comparison of visual core object recognition performance between humans and feedforward neural networks in a classification learning paradigm on an ImageNet data set. For this purpose, human participants (n = 65) competed in an online experiment against different feedforward DCNNs. The designed approach based on a typical learning process of seven different monkey categories included a training and validation phase with natural examples, as well as a testing phase with novel, unexperienced shape and color manipulations. Analyses of accuracy revealed that humans not only outperform DCNNs on all conditions, but also display significantly greater robustness towards shape and most notably color alterations. Furthermore, a precise examination of behavioral patterns highlights these findings by revealing independent classification errors between the groups. The obtained results show that humans contrast strongly with artificial feedforward architectures when it comes to visual core object recognition of manipulated images. In general, these findings are in line with a growing body of literature, that hints towards recurrence as a crucial factor for adequate generalization abilities.

Science communication

2023

1. Modellierung der biologischen Gesichtswahrnehmung mit Künstlicher Intelligenz

   van Dyck, L. E., and Dobs, K.

   DER AUGENSPIEGEL