120 minutes
4–6 pm
Technical
Guest lecture
CITS5553 · Data Science Capstone Project Master of Data Science · The University of Western Australia
Multimodal LLMs and VLA
From partial signals to systems that act
Why can a current AI system make a credible start on a gravitational model of the Solar System, yet struggle to pass someone a red mug? This lecture uses that reversal to introduce multimodality, develops contrastive learning from first principles, and follows learned representations into vision–language models, embodied policies, and predictive world models.