Malin Styrnal

I am a PhD student at the Indivisual Lab at the Justus Liebig University Giessen under the supervision of Prof. Ben de Haas.

I want to understand how our visual system works, particularly how our brain recognizes and understands the objects our eyes see. Objects play a fundamental role in our daily lives, and the ability to rapidly and effortlessly recognize and comprehend them is crucial for effective interaction with our environment. But how is our brain able to do that and how are objects represented in the brain?

During my PhD I initially investigated the comparison of similarity measures and am now exploring individual differences in gaze behaviour.

CV

Research and Work Experience

Indivisual Lab (Prof. de Haas)

Justus Liebig University Giessen

since Feb. 2026

PhD Candidate

Youth Migration Services and Migration Counseling for Adult Immigrants

IB Südwest

since Apr. 2025

Social Worker

Computational Cognitive Neuroscience and Quantitative Psychiatry (Prof. Hebart)

Justus Liebig University Giessen

Sep. 2023 – Jan. 2026

PhD Candidate

Max Planck Research Group Vision and Computational Cognition

Max Planck Institute for Human Cognitive and Brain Sciences

since May 2023

Guest Scientist

General Psychology and Methodology (Prof. Carbon)

UNiVERSiTY OF BAMBERG

Aug. 2019 ‑ Jun. 2023

Student Research Assistant

Icelandic Vision Lab (Prof. Kristjánsson)

University of Iceland

Aug. 2022 ‑ Oct. 2022

Research Internship

Leibniz Institute for Educational Trajectories

Bamberg

Aug. 2018 ‑ Mar. 2019

Student Research Assistant

Education

MSc in Psychology

UNiVERSiTY OF BAMBERG

Oct. 2020 ‑ Mar. 2023

Thesis: Kinetic‑depth effect multistability persists when attention is distracted by an attention‑demanding RSVP task.

BSc in Psychology

UNiVERSiTY OF BAMBERG

Oct. 2017 ‑ Sep. 2020

Thesis: Perceptual history alters shape (not just the mean) of dominance phase distribution.

Publications

Paper

How Much Variance Does Your Model Explain? A Clarifying Note on the Use of Split-Half Reliability for Computing Noise Ceilings

van Bree, S., Styrnal, M., Hebart, M.N. (2025).
PsyArXiv.

KEYWORDS: computational model, correlation, encoding model, explained variance, noise ceiling, reliability, Spearman-Brown

Noise ceilings estimated from a dataset’s split-half reliability offer a powerful way to quantify how much variance a model can in principle explain given the noise in the dataset, allowing researchers to assess model performance relative to an upper bound. In this work, we caution against a common pitfall in this approach to estimating noise ceilings. Specifically, even though the split-half reliability is expressed as a correlation coefficient, it reflects the maximum explained variance of a perfect model, not the maximum correlation. This subtle misinterpretation leads to artificially lower noise ceilings and, as a consequence, may inflate how close models appear to be to the noise ceiling. A systematic literature analysis suggests that this overly per- missive ceiling is the most prevalent interpretation of noise ceilings estimated through split-half reliability. The purpose of this work is to explain when the mistake happens, why it happens, what its consequences are, and how to avoid it. Toward this end, we offer a general explanation showing how split-half reliabilities relate to the performance of a maximally predictive model, supplemented by simulations, and mathematical derivations. Overall, this clarifying piece is meant to help researchers better understand the statistical underpinnings of noise ceilings and support more consistent reporting across studies.

Two routes to a target: Visual priming for direct and indirect attentional sets

Pastukhov, A, Styrnal, M, Carbon, C.C. & Kristjánsson, Á. (2025).
Memory & Cognition.

KEYWORDS: repetition priming, attentional sets, conjunction search, visual search

To find an item of interest among candidate objects we are directed by attentional sets that reflect our expectations and intentions and may also vary by whether items should be attended or ignored. We investigated how different attentional sets influence target search and the effect of prior experience on these attentional sets. Our participants had to identify a target object given a set of objects that either contained the target itself (direct attentional set) or contained only cues that defined the target by exclusion (indirect attentional set). We found that response times were significantly slower for indirect attentional sets and when sets were mixed within blocks. To analyze the impact of attentional sets on priming, we fitted behavioral time series using multiple dynamic ideal observer models based on a first-order memory mechanism with three consecutive stages: set identification (direct vs. indirect), target identification (based on set cues), and response. The different models involved different assumptions about each stage, and we compared them via information criterion to identify mechanisms that consistently lead to good expected out-of-sample performance. We found strong repetition priming when both set and target were repeated. For direct attentional set, repetition priming was consistent with a first-order memory mechanism that tracks objects and colors likely connected to feature-specific neural mechanisms and frontoparietal attention network. In contrast, the processing of indirect attentional sets relies on qualitatively different mechanisms and search strategies than conventional visual search, likely related to neural networks involved in task switching and generation of attentional set.

When a bank becomes a bank, and a bank is the bank but not the bank: Multistability of homonyms’ meaning

Styrnal, M., Carbon, C. C. & Pastukhov, A. (2023).
i-Perception.

KEYWORDS: rivalry/bistability, audition, cognition, perception

Perceptual multistability is well-known and mostly visually demonstrated: Common examples are Necker’s cube or Rubin’s face-vase that produce qualitatively different percepts continuously oscillating between the solutions despite physically stable stimuli. We lack knowledge about similar phenomena in other domains, for instance in linguistics, where we are faced with homonyms that create multistability of cognitive semantics, differently assigned meanings of identical words. Our participants listened to repeated presentations of homonyms for which two or even three meanings could be assigned, and they reported the dominant meaning perceived at a certain point in time. Results showed that most participants experienced multistability of meaning for homonyms, with semiperiodic changes in dominant meaning similar to multistabity in perception. These findings suggest that multistability is a general property of the brain’s neural architecture that resolves ambiguity irrespective of the level of representation.

History-dependent changes to distribution of dominance phases in multistable perception

Pastukhov, A., Styrnal, M. & Carbon, C. C. (2023).
Journal of Vision.

KEYWORDS: multistable perception, serial dependence, history dependence, binocular rivalry, necker cube, kinetic-depth effect

Multistability – spontaneous switches of perception when viewing a stimulus compatible with several percepts – is often characterized by the distribution of durations of dominance phases. For continuous viewing conditions, these distributions are similar for various multistable displays and share two characteristic features: a Gamma-like distribution shape and dependence of dominance durations on the perceptual history. Both properties depend on a balance between self-adaptation (also conceptualized as a weakening stability prior) and noise. Prior experimental work and simulations that systematically manipulated displays showed that faster self-adaptation leads to a more „normal-like“ distribution and, typically, to more regular dominance durations. We used a leaky integrator approach to estimate accumulated differences in self-adaptation between competing representations and used it as a predictor when fitting two parameters of a Gamma distribution independently. We confirmed earlier work showing that larger differences in self-adaptation led to a more „normal-like“ distribution suggesting similar mechanisms that rely on the balance between self-adaptation and noise. However, these larger differences led to less regular dominance phases suggesting that longer times required for recovery from adaptation give noise more chances to induce a spontaneous switch. Our results also remind us that individual dominance phases are not „independent and identically distributed.“

Conference Talks

Kinetic-depth effect multistability persists when attention is distracted by an attention-demanding RSVP task

Styrnal, M., Pastukhov, A. & Carbon, C. C.
ECVP 2022.

KEYWORDS: multistable perception, attention, kinetic-depth effect

Does multistable perception require attention? The answer is “yes” for binocular rivalry but “probably not” for other multistable displays. We re-examined this question in two experiments using kinetic-depth effect (KDE) displays where participants viewed multiple ambiguously rotating objects (two in Experiment 1, three in Experiment 2, randomly but evenly distributed along the circle), the initial direction of rotation was randomly biased via size/distance cues. After the bias- ing cues were gradually removed, axes of rotation rotated from vertical to horizontal over a two-second interval and all but one object disappeared with participants reporting on its direc- tion of rotation. Orthogonal orientation of initial and final axes of rotation ensured that initial and final direction of rotation could be correlated only if multistability persisted during axes rotation. As a control, we included a “jump” condition with rotation axes abruptly changing their orientation. In Experiment 1, we manipulated attention via validity cues with different colors indicating the probability of an object being a target (50%, 67%, 83%, 100%). Here, we found that persistence of perceptual dominance was not affected by validity cues and was immune to any difference in allocation of spatial attention. In Experiment 2 participants performed an attention-demanding RSVP task during the axes rotation interval. We found that although the persistence of perceptual dominance was weaker during the inattention condition, it was significantly higher than that for the jump condition. Taken together this indicates that the multistability of KDE persists without attention.

Multistability: Perceptual history alters shape (not just the mean) of dominance phase distribution

Styrnal, M., Pastukhov, A. & Carbon, C. C.
ECVP 2021.

KEYWORDS: multistable perception, serial dependence, history dependence, binocular rivalry, necker cube, kinetic-depth effect

Very different multistable displays produce time series of dominance phases that are qualitatively and quantitatively similar, including the characteristic right-skewed distribu- tion and subtle but consistent dependence on prior percep- tual experience. Prior work examined how recent perceptual experience alters the mean of the distribution. However, for an asymmetric right-skewed distribution such as Gamma that is commonly used to fit time-series for multistable displays, the same change in the mean can come about through different changes in one or both its parameters. We investigated this by employing a Gamma distribution with both parameters being (under the log link) linearly dependent on prior perceptual experience and further factors (age of participants, contrast). Five data- sets were fitted with a multilevel Bayesian model that included different multistable displays with constant strength (binocular rivalry, kinetic-depth effect, Necker cube), binocular rivalry with varying contrast, and developmental data. For all data sets, we found that only the shape parameter depended on a prior perceptual history being compatible with the idea of an interplay between adaptation and noise. Specifically, longer accumulated perceptual history assumed to reflect higher levels of adaptation resulted in a more “normal-like” distribution associated with an adaptation-driven oscillatory regime. Conversely, lower accumulated perceptual history (as a proxy for adaptation) produced a more “exponential-like” distribution consistent with noise-driven alternations. Our results provide useful constraints for computational models of multistability, as, for example, they cannot be easily reproduced by a classic spiking neuron model by Laing and Chow.

Conference Posters

The similarity of similarity tasks: Comparing eight different measures of similarity.

Styrnal, M., Kaniuth, P., Stoinski, L. & Hebart, M. N.
ECVP 2025.

KEYWORDS: similarity tasks, object similarity, object recognition, methods

Similarity tasks are widely used in the field of (cognitive) psychology and beyond. However, when deciding on an appropriate measure of similarity, one needs to choose from a wide range of different tasks with various estimates of similarity, such as odd-one-out judgments, response time tasks, distance-based measures, etc. So, which task is most reliable, cost efficient and best suited for the aims of our study? To address these questions, we compared eight different similarity measures: (1) triplet odd-one-out judgments, (2) sequential forced choice similarity judgments, (3) speeded visual search, (4) speeded same-different judgments, (5) pairwise ratings, (6) single and (7) multiple object arrangement and (8) pile sorting. We collected data for all tasks using online crowdsourcing and three stimulus sets with decreasing complexity: a set of natural, colored images from the THINGS database, a set of grayscale objects cropped from the image background and a set of black shapes. We analyzed the similarity estimates between tasks and their alignment with representations from several VGG19 layers. Our findings revealed three groups of tasks: (a) tasks that primarily capture visual features (response time tasks), (b) tasks that primarily capture semantic features (sorting tasks) and (c) tasks that capture both visual and semantic features (choice tasks). The extent of differences in similarity structure between groups depends on stimulus complexity, with more complex stimuli resulting in larger differences. Within each group, we identified tasks that are more reliable and efficient than others. Thus, when choosing a similarity task researchers should take into account which kind of similarity they intend to study, the type of stimuli they plan on using as well as the reliability and efficiency of the tasks. Currently, we are conducting an EEG study to investigate how the behavioral responses from the tasks relate to brain activity.

show full abstract

show less

PDF

What do similarity tasks actually measure? A systematic comparison of eight tasks.

Styrnal, M., Kaniuth, P., Stoinski, L. & Hebart, M. N.
ECVP 2024.

KEYWORDS: similarity tasks, object similarity, object recognition, methods

Perceived similarity judgments play a critical role in the measurement of mental representations, offering a way to understand how we represent what we see. Numerous similarity tasks are used routinely in research on object perception that estimate perceived similarity with a broad range of behavioral techniques. Despite decades of their use, the degree to which these tasks are measuring the same construct and their relative reliability and external validity are largely unknown, leaving a significant gap in the literature concerning a comprehensive and systematic comparison of commonly used similarity measures. To address this issue, our aim is to systematically compare different similarity measures using eight common tasks: 1) pairwise ratings, 2) single and 3) multiple object arrangements, 4) triplet odd-one-out judgments, 5) pile sorting, 6) a speeded visual search, 7) speeded same-different judgments, and 8) sequential forced choice similarity judgments (SFCS). The latter is a novel and, potentially, more time-efficient alternative to the traditional odd-one-out task. We will compare these tasks using their inter-observer and intra-observer reliability, efficiency (data acquisition speed) and agreement with neuroimaging data (fMRI and EEG). The goals of this project are to systematically compare commonly used similarity measures, to assess the role of the task on mental representations of visually perceived objects, to understand similarities and differences between tasks and to identify which tasks are most suited for specific contexts, as well as different types of experiments and research questions.

show full abstract

show less

PDF

Geistesblitz: Repetition priming of indirect attentional sets

Styrnal, M., Carbon, C. C., Pastukhov, A. & Kristjánsson, Á.
ECVP 2023.

KEYWORDS: target identification, visual search, priming, attention, attentional set

During visual search for a unique target repeating the target features leads to faster search (repetition priming). But what happens if the features do not include the target and the target cue is negative, thus defined via exclusion? In other words, what happens if participants must figure out which stimuli are not the target to find the one that is? To address this, we employed two attentional sets for specifying targets: direct and indirect sets. In the former, the target is present on the screen, while in the latter, the target is cued via exclusion. We wanted to understand if not only the stimulus but also the way participants have to search for a target (the attentional set) is primed. We conducted three experiments based on the board game “Geistesblitz (Ghost Blitz)”. In each trial, we presented several objects in different colours on the right side of the screen and a direct or indirect attentional set on the left side. Participants had to find and select the target on the right as fast as possible based on the cues in the attentional set. In the direct set, one of the objects matched exactly one of the potential targets while in indirect sets none of the objects matched exactly and participants had to find the one object that was not one of the objects in the set and not in one of the colours in the set. There was always only one correct target. We developed a series of ideal observer models that suggest repetition priming of the objects occurring independently for the two attentional sets. In addition, colour priming occurs independently of object repetition priming. Finally, we found only weak evidence for priming of the set itself with a differential time-penalty for identifying the set explaining response times equally well.

Bistable photos do not pop out but provoke a prolonged inspection

Styrnal, M., Billing, K., Pastukhov, A. & Carbon, C. C.
ECVP 2019.

KEYWORDS: multistable perception, gaze, ambiguity, visual perception

show full abstract

show less

Link

Blog

Tutorial: How to use Connect to recruit participants for online studies

Styrnal, M. (2024).

KEYWORDS: online studies, participant recruitment, Connect, CloudResearch

Connect is CloudResearch’s participant recruitment and crowdsourcing platform that offers high quality data for comparably low costs.
This tutorial covers how recruiting participants for online studies works and how to use the survey and HTML templates. For the HTML template practical examples on micro-tasks and how to integrate Pavlovia studies are provided.
The tutorial also offers some tips and tricks on how to deal with cheaters on Connect but also in general when running online studies.