Eye-Tracking · Cognitive Neuroscience · Affective Computing · HRI
I am a Postdoctoral Researcher at the Department of Applied Mathematics and Computer Science, Technical University of Denmark (DTU), working on the Reading the Reader project funded by the Novo Nordisk Foundation, in collaboration with Assoc. Prof. Per Bækgaard.
My research sits at the intersection of Cognitive Neuroscience, UX/UI Design, Human–Machine Interaction (HMI), Human–Robot Interaction (HRI), and Artificial Intelligence — using eye-tracking and psychophysics to understand how people read, perceive, and interact with information.
My research spans cognitive science, affective computing, multimodal signal processing, and human-centred AI — developing computational methods that place human perception, attention, and emotion at the centre of intelligent system design.
Investigating how visual attention mechanisms operate using advanced eye-tracking technology and computational models based on Bundesen's Theory of Visual Attention (TVA). Leveraging Tobii eye-trackers to capture gaze trajectories, fixation dynamics, and Task-Evoked Pupillary Responses (TEPR) — then linking these signals to higher-level cognitive processes such as reading fluency, perceptual load, and attentional capacity.
Unifying Bundesen's TVA with Bayesian hierarchical modelling to quantify individual differences in visual processing. Per-participant estimation of attentional parameters (visual processing speed C, storage capacity K, threshold t₀) using MCMC in PyMC. Psychophysical experiments examine how typography, font legibility, and stimulus properties modulate perception — with posterior difference distributions, Growth Curve Analysis (GCA), and LOO-CV for robust inference.
Designing signal processing pipelines that fuse heterogeneous physiological streams — eye-tracking, pupillometry, EEG, and facial action units — into unified representations of cognitive and affective state. Research covers time-series alignment, artifact rejection, feature engineering, and the development of robust multimodal models for real-world sensing contexts including reading, workplace monitoring, and HRI. Central theme of the MSPARC workshop at IEEE ICASSP 2026.
Building emotionally intelligent machines that can perceive, interpret, and respond to human affective states. Facial expression recognition (FER) with dimensional valence/arousal modelling, adapted for socially assistive robots supporting citizens with Traumatic Brain Injury (TBI). Includes work on spontaneous affect in naturalistic environments (EU Horizon 2020 WorkingAge project, Cambridge AFAR Lab), where systems must be robust to unconstrained lighting, occlusion, and real-world variability.
Developing socially assistive and therapeutic robotic systems that support cognitive and physical rehabilitation for individuals with neurological conditions. PhD research focused on deploying facial emotion recognition within robot-assisted therapy for citizens with Traumatic Brain Injury (TBI) — enabling robots to perceive patient affective states and adapt therapeutic interactions in real time. Research intersects clinical HRI, affective sensing, and the ethical deployment of autonomous systems in healthcare and assistive technology environments.
Studying how humans interact with intelligent systems — and using that understanding to build interfaces that adapt in real time to the user's cognitive and emotional state. Research spans gaze-contingent reading aids, attention-aware document rendering, and closed-loop interfaces that leverage eye-tracking and pupillometry to reduce cognitive load and personalise content presentation. Bridges HCI methodology with physiological computing and AI-driven adaptation.
Applying deep neural architectures to perception and cognition problems: convolutional networks and transformers for facial action unit detection, expression recognition, and affect estimation from video streams. Research emphasises domain generalisation, transfer learning across affective datasets, and the design of lightweight models suitable for real-time deployment in embedded and robotic systems. Also encompasses BERT-based sequence models applied to gaze reading data for language identification.
Advocating for AI systems designed with human needs, values, and diversity at their core. This thread runs through all my work — from building models that explain their inferences in terms of measurable cognitive parameters, to designing experiments with genuine ecological validity, to ensuring assistive technologies serve populations with neurological and cognitive differences. Engages with transparency, fairness, and the responsible deployment of AI in health, education, and workplace settings.
Combining natural language processing with gaze behavioural data to study reading and language processing across populations. BERT-based classifiers applied to fixation sequences for native language identification from reading patterns (NLIR-BERT), evaluated on the MECO-L2 multilingual corpus spanning 13 languages. Explores how gaze reflects lexical access, syntactic processing, and L2 reading proficiency — with Monte Carlo simulation for robust small-sample statistical inference.
Peer-reviewed contributions to eye-tracking, cognitive science, and human-computer interaction. View full list on Google Scholar →
ACM Symposium on Eye Tracking Research and Applications (ETRA 2025), Tokyo, Japan · May 2025 · Article 78 · Open Access CC BY 4.0
Introduces context preservation — a gaze-based mechanism that helps readers resume reading after typographic adaptations. A 22-participant within-subjects experiment compared four intervention designs (Popup, Undo, Notification, Gradual). The Gradual intervention achieved the lowest Reading-Resume Time and highest participant satisfaction.
ACM Symposium on Eye Tracking Research and Applications (ETRA 2025), Tokyo, Japan · May 2025 · Article 113 · Open Access CC BY 4.0
First eye-tracking investigation comparing cognitive reading behaviour between LLM-generated and human-authored texts. Gaze data from 13 participants analysed using the I2MC fixation detection algorithm. Significant differences in fixation characteristics, pupil dilation, and reading speed identified between AI and human text.
IEEE 7th International Conference on Cybernetics, Cognition and Machine Learning Applications (ICCCMLA 2025), 1–2 November 2025 · In Press
First BERT-based classifier for Native Language Identification from Reading (NLIR) using gaze features from the MECO-L2 Wave 1 corpus. Monte Carlo simulation with 20 randomised splits for robust small-sample evaluation. BERT produces linguistically more meaningful language clusters than logistic regression despite comparable accuracy.
arXiv preprint · WorkingAge / EU Horizon 2020 Project · November 2021 · arXiv:2111.11862
A system for continuous valence/arousal affect inference from facial cues in naturalistic work-like environments, developed within the EU H2020 WorkingAge project. Addresses challenges of subtle workplace expressions, variable illumination, and unobtrusive sensing — demonstrating that dimensional models outperform categorical classification for in-the-wild occupational affective computing.
Organizing and leading advanced technical workshops at international venues.
A full-day workshop at IEEE ICASSP 2026 bringing together researchers working at the intersection of multimodal signal processing, eye-tracking, pupillometry, and cognitive science. The workshop covers advanced methods for analysing attentional processes through physiological signals, with a focus on real-world applications in reading, learning, and human–computer interaction.
A trajectory spanning four countries, combining basic cognitive research with applied AI and human-centred computing.
Working on the Reading the Reader project (Novo Nordisk Foundation, grant NF22OC0076877) with Assoc. Prof. Per Bækgaard. Developing eye-tracking and Bayesian methods to understand typographic effects on reading and deploy adaptive, AI-driven reading interfaces. Collaborating with the Royal Danish Academy — Architecture, Design, Conservation.
Contributed to the WorkingAge project (EU Horizon 2020, grant No. 826232), focusing on affective analysis of workers using computer vision, multimodal sensing, and deep learning to promote healthy habits and improved wellbeing in professional settings.
Doctoral research on Facial Emotion Recognition for Citizens with Traumatic Brain Injury for Therapeutic Robot Interaction, conducted jointly within the Visual Analysis & Perception (VAP) and Human–Robot Interaction (HRI) Labs.
Graduated with a Master's degree in Electrical & Electronics Engineering. Awarded the Global Leadership Certificate (GLC) for organising seminars, events, and talks under the Global Leadership Programme (GLP).
Science communication pieces written during my PostDoc at the AFAR Lab, University of Cambridge, published on the WorkingAge project blog (EU Horizon 2020).
Reporting on work presented at the "Human-Machine Partnerships in the Future of Work" workshop at CSCW'21. Describes a facial affect analysis system for naturalistic work settings that models workers' emotional states using valence and arousal dimensions — showing that pre-trained models on naturalistic datasets generalise well to workplace affect.
The second part of the RED-AI series covers practical techniques for training large machine learning models more efficiently — addressing the growing carbon footprint and accessibility barriers raised in Part I, and offering concrete methods to make deep learning research more sustainable and open.
Introduces the concept of "RED AI" — the trend of prioritising accuracy over computational efficiency. Examines how the compute cost of state-of-the-art models grew 300,000× between AlexNet (2012) and AlphaZero (2017), creating carbon footprint concerns and raising the entry bar for academic researchers.
Covers the CARE 2020 workshop ("Pattern Recognition for positive teChnology And eldeRly wEllbeing") at ICPR2020. Discusses computational models for automatically evaluating the emotional wellbeing of elderly people using non-verbal social signals — facial expressions, body language, and gestures — in ecological, unconstrained settings.
Co-authored with Dr. Hatice Gunes (Cambridge). Reports on the European Beyondwork2020 conference (Bonn, Germany), exploring how cognitive training via virtual reality and affective adaptation can shape healthier, more supportive workplaces in the post-COVID era.
Supervising student projects at DTU Compute, Department of Applied Mathematics and Computer Science, within the Reading the Reader research group. Projects span eye-tracking, adaptive reading interfaces, typography, cognitive AI, and HCI — at BSc, MSc, and Diploma level.
Modelling emotional and behavioural states in occupational and real-world settings.
Designing socially intelligent robots that perceive and respond to human affect.
Brain-inspired computation for perception, reading, and decision-making.
Facial expression recognition, action recognition, and multimodal sensing.
Typography, legibility, adaptive interfaces, and eye-tracking for literacy support.
Hierarchical models, MCMC inference, and probabilistic analysis of cognitive data.
Selected recognitions across research, leadership, and innovation.
Developing AI-driven adaptive reading interfaces using eye-tracking and biometric sensors, in collaboration with the Royal Danish Academy and DTU Compute's Cognitive Systems Section.
Affective analysis of workers to promote healthy habits, improve working conditions, and support well-being in professional and daily living contexts. Hosted at AFAR Lab, Cambridge.
Open to research collaborations, supervision discussions, invited talks, and questions about my work.