High-level event segmentation: towards a systematic model for mixed-methods research in AI and Visual perception

The talk presented a high-level event segmentation framework aimed at developing a systematic model for mixed-methods research in AI and visual perception. It drew on multiple research projects examining how visuoauditory cues — including gaze, speech, motion, and hand actions — guide visual attention during passive observation of naturalistic human interactions. Central emphasis was placed on integrating a structured event model with eye-tracking data to study attention within ecologically valid interaction scenarios, and to characterize how these cues operate both independently and in combination. The findings demonstrated robust intra- and cross-modal attentional effects, highlighting how perceptual and interactional signals jointly shape attentional allocation in real-world contexts. The proposed framework illustrates a principled approach for linking event structure, multimodal behavior, and attention, with implications for cognitive modeling, human-centered AI, and the design of interactive systems.

Talk-Workshop

Artificial and Human Intelligence Workshop at the 24th European Conference on Artificial Intelligence, Santiago de Compostela, Spain, Sept 2020 (Workshop ECAI 2020)*