AGCM is a concept-based framework that links affective predictions to predefined behavioural concepts and their spatial evidence. It jointly learns concept representations and task predictions within an end-to-end architecture, bridging concept-level interpretability with predictive modelling. The model reveals which concepts drive each decision and where the supporting visual evidence is located.

Demo & pretrained models: https://github.com/xinyuli-uofg/agcm.
Selected Publication: Li, Xinyu and Mahmoud, Marwa (2026) AGCM: Attention-Guided Concept-based Model for Interpretable Affective Computing Applications. In: 20th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2026)

ConceptMamba is an interpretable temporal model that learns affective concepts as the language for explaining predictions in mental-health applications. It combines concept-based modelling with Mamba-based temporal modelling, capturing how interpretable affective cues evolve over time. The framework provides concept-level explanations alongside predictions for applications such as depression and psychological distress analysis, enabling more transparent modelling of complex temporal behaviour.

Selected Publication: Li, Xinyu and Mahmoud, Marwa (2026)  ConceptMamba: Interpretable Temporal Model for Mental Health Applications. In: 2026 International Conference on Affective Computing and Intelligent Interaction (ACII 2026)