My caption 😄
Affective Computing
Human affect is dynamic, multimodal, and highly variable. Emotions are expressed through facial behaviour, voice, language, body movement, physiology, and context. Our research develops computational methods to recognize, represent, model, and predict human affective behaviour across these different signals.
We train multimodal AI models, including large language models (LLMs) and vision-language models (VLMs), to predict and reason about human emotion from combinations of language, images, video, speech, and behavioural signals. We also develop specialized machine-learning models to investigate how affective information is represented across modalities.
A major focus of our work is emotional dynamics. Rather than asking only “What emotion is this person expressing?”, we study how affect changes over time—including patterns of intensity, variability, stability, inertia, and relationships between emotional states. This allows us to model emotion as an evolving process rather than a collection of static labels.
Alongside building predictive models, we use theory-driven computational approaches to investigate what these models can tell us about human affective behaviour. We examine individual differences, multimodal cues, and model generalization across datasets and contexts.
Our research spans
Emotion Recognition & Prediction
AI models that infer affective states from facial, vocal, linguistic, and behavioural signals.
LLMs & VLMs for Affect
Adapting and evaluating foundation models for understanding and predicting emotions across language, vision, video, and other modalities.
Emotional Dynamics
Modeling how affect changes and unfolds over time, including intensity, variability, stability, inertia, and interactions between emotional states.
Multimodal Affect
Combining information across face, voice, language, physiology, and behaviour to build richer representations of emotion.
Individual Differences
Understanding variability in how people experience, express, and communicate emotion.
Theory-Driven AI
Using computational models not only to build better systems, but also to test hypotheses about human affective behaviour.
These models support applications in health and wellbeing, human–AI interaction, conversational agents, education, and other settings where understanding human affect is important.
