PhD Studentship: Simulating Person-specific Cognitive Processes from Expressive Behaviours for Mental Health Assessment. Funded PhD in Computer Science

University of Exeter; Devon and Torbay; On-site; GBP 21,000 - 23,000; Full time

Job summary

The University of Exeter is offering a funded PhD studentship in the Department of Computer Science, starting 21 September 2026 or as soon as possible thereafter. The project develops personalised AI models for computational mental health using multimodal data and advanced neural architectures.

It aims to infer mental health states from expressive behaviours with emphasis on fairness, privacy, and explainability.

Qualifications

• Strong programming skills are essential.
• Background in computer science, AI or ML is preferred.
• Experience in affective computing or multimodal signal processing is advantageous.

Responsibilities

• Develop multimodal models from facial expressions, speech, gaze and movement.
• Encode cognition using graph- and transformer-based methods.
• Train and evaluate models on publicly available and ethically collected datasets.
• Investigate fairness, privacy, uncertainty and explainability for responsible AI.

Skills

Programming skills

Machine learning

Affective computing

Multimodal signals

Education

Background in Computer Science/AI/ML

Job description

The University of Exeter’s Department of Computer Science invites applications for a funded PhD studentship to commence on 21 September 2026, or as soon as possible thereafter. For eligible students, the studentship will cover Home or International tuition fees plus an annual tax-free stipend of at least £21,805 for 3.5 years full-time, or pro rata for part-time study. The student will be based in the Department of Computer Science, Faculty of Environment, Science and Economy, at the Streatham Campus. The studentship is funded by the University Startup Package and supported by the University and the Department of Computer Science.

Mental health conditions such as depression, anxiety and bipolar disorder are commonly assessed through self-report questionnaires and clinical interviews. While valuable, these methods can be subjective, time-consuming and difficult to scale. Meanwhile, mental health states may also be reflected in how people express, perceive and respond to the world through facial behaviour, speech, gaze, affective dynamics and interaction patterns.

This PhD project will develop personalised AI models for computational mental health by simulating the latent cognitive processes underlying individual expressive behaviours. The project builds on recent work showing that person-specific cognition can be computationally approximated by learning personalised neural architectures or model weights that reproduce an individual’s facial reactions to audio-visual stimuli. Rather than treating the learned model as a simple behaviour classifier, this project will use it as a representation of an individual’s internal cognitive-response mechanism for downstream mental health inference. The central hypothesis is that depression, anxiety and bipolar disorder are associated with distinctive personalised cognitive-response mechanisms that can be inferred more robustly from expressive behaviours than from surface-level behavioural features alone. Instead of directly predicting diagnostic labels from facial or speech features, the PhD will first learn an individualised cognitive simulator that models how a person generates expressive responses under emotional, social or conversational contexts. The simulated cognition will then be used to infer mental health states, symptom severity and temporal changes.

The project will pursue four objectives: developing multimodal models from facial expressions, speech, head movement, gaze and interaction dynamics; encoding simulated cognition using graph- and transformer-based methods; training and evaluating models on publicly available and ethically collected clinical or sub-clinical datasets; and investigating fairness, privacy, uncertainty estimation and explainability so that predictions are reliable and used as decision-support rather than automated diagnosis. The student will work with deep learning, affective computing, multimodal signal processing, graph neural networks, hypernetworks, temporal modelling and responsible AI. Expected outputs include personalised cognition simulation algorithms, benchmark evaluations against direct behaviour-to-label baselines, interpretable markers of cognitive-affective dysfunction, and prototypes for non-invasive mental health monitoring. Candidates should have a background in computer science, AI, machine learning, affective computing, computational psychology or related areas. Strong programming skills are essential.

UK or International tuition fees and an annual tax-free stipend of at least the UKRI minimum stipend