BEGIN:VCALENDAR PRODID:-//Columba Systems Ltd//NONSGML CPNG/SpringViewer/ICal Output/3.3- M3//EN VERSION:2.0 CALSCALE:GREGORIAN METHOD:PUBLISH BEGIN:VEVENT DTSTAMP:20261006T101750Z DTSTART:20261007T121500Z DTEND:20261007T131500Z SUMMARY:Departmental Seminar: “Trustworthy AI for Subtle Visual Signals” – Dr Xinqi Fan\, Manchester Metropolitan University UID:{http://www.columbasystems.com/customers/uom/gpp/eventid/}r9z-muwhsm0 x-rcqu7t DESCRIPTION:Abstract\n\nSome of the informative visual evidence appears a s small changes that are easy to overlook. Learning from these subtle si gnals remains challenging for AI systems. This talk presents our researc h on trustworthy AI for recognising subtle visual signals across facial and medical image understanding. First\, facial micro-expressions are br ief and subtle movements that may provide cues to underlying affective s tates. We investigate robust representation learning through self-superv ised motion learning to capture their fine-grained dynamics. We further demonstrate that the learned representations can support both micro-expr ession analysis and video generation. Second\, we turn to medical imagin g for ulcerative colitis\, where small differences in mucosal appearance can affect disease assessment. To ground predictions in clinical knowle dge\, we develop an endoscopic-informed spiral-scanning strategy for sta te-space models that captures meaningful visual patterns. We also introd uce a mixture of low-rank vision-language experts to incorporate clinica l concepts into model training and reasoning. Finally\, small difference s between training and deployment conditions can reduce model performanc e. We address this distribution shift through retrieval-augmented test-t ime adaptation and reflective multi-agent reasoning\, enabling AI system s to use new evidence and self-improve their predictions without convent ional retraining.\n STATUS:TENTATIVE TRANSP:TRANSPARENT CLASS:PUBLIC LOCATION:Kilburn_TH 1.3\, Kilburn Building\, Manchester END:VEVENT END:VCALENDAR