BMVA Symposium on Interpretability

Topics include what comes next for interpretability in vision, mechanistic interpretability and circuit discovery, post-hoc analysis after training, interpretable-by-design methods, language-based and multimodal techniques, white-box versus black-box approaches, making foundation models interpretable, datasets and evaluation metrics, understanding representations for explainable models, model intervention, robustness and scalability, and real-world applications. Chaired by Davide Moltisanti and Michael Wray, the tentative programme (09:00-17:00) includes invited talks by Oisin Mac Aodha, Elisa Ricci and Stephan Alaniz, accepted talks, poster sessions and an interactive closing session. Presentations may cover published or ongoing work as talks or posters; expressions of interest to present are due by 26 November 2026. The event is in person only; registration costs £30 including lunch and refreshments.