Mask R-CNN author Georgia Gkioxari argues closed-loop self-improving perception is AI's next leap

CSProfKGD · x · 2026-09-18

Caltech assistant professor and Mask R-CNN co-author Georgia Gkioxari presents "Towards Self-Improving Machine Perception" at Penn's GRASP lab.

Her core argument: AI's next leap will come not only from larger models but from closed-loop learning systems that generate, critique, and re-curate their own training data. The talk covers her recent work on 3D perception, semantic grounding, and spatial reasoning, and how self-improving learning could enable more capable world models, more reliable perception, and systems that learn in domains with scarce labeled data.

Gkioxari is a Marr Prize winner, PAMI Young Researcher Award recipient, and a 2025 Packard Fellow.

Original post →

More from Embodied

Embodied channel →