Yifan Luo (KU Leuven) – Incorporating Domain-General Capacities into Domain-Specific Models
(co-PI: Hans Op de Beeck, Andrea Costantino)
get in touch: yifan.luo (at) kuleuven (.) be

Contemporary AI is superb at perception. Human vision, by contrast, couples fast recognition with slow, reflective control: we gauge uncertainty, revise initial impressions, and sometimes decide not to decide. I’m working to weave this metacognitive fabric into artificial vision. The idea is that intelligence is not just about solving tasks, but also about knowing when to pause, doubt, and re-evaluate. We aim to extend computational models of human vision (e.g., deep convolutional neural networks) beyond narrow, domain specific strengths, to bear domain-general capacities, like metacognition that make human cognition flexible and adaptive, and finally to build uncertainty-aware, metacognitive AI inspired by the human visual system.