Xia Xu defended his thesis in the group of Jochen Triesch on December 10. He studied how infant-inspired learning can lead to the next level of developmental AI.

While infants are remarkably efficient at learning through observation and interaction, modern deep learning models typically require massive amounts of data and computational resources. To bridge this gap, Xu's work at the Triesch Lab charted a computational path from passive observation to active, causal agency.

Through several publications, he explored how to combine modern techniques like Contrastive Learning with the infant-inspired "temporal coherence principle." The group also investigated how to make passive models aware of their actions through equivariant representations, and modeled infant behavior through the lens of "sense of agency" using a novel causal mechanism. These stages equip artificial learners with cognitive tools that serve as a solid foundation for the next level of developmental AI.

Xu is currently seeking opportunities in academia and hope to continue exploring exciting ideas at the intersection of biology, psychology, and AI.