Post by Helmholtz AI

"It sounds a bit counterintuitive," but according to Pierre Gentine (Columbia University), AI models that encode physics into latent spaces, rather than tracking a system's real state directly, can actually predict complex physical systems better, especially over the long term. That is the idea at the heart of his #HAICON26 keynote talk, "Lost (and found) in latent land: applications to weather and climate." Even as AI reaches unprecedented accuracy in weather and climate modeling, Gentine argues it's not always clear how much real understanding these models generate. His work shows that carefully chosen latent spaces can change that, surfacing new insight into the water and carbon cycle, atmospheric processes like convection, and even the chaotic behavior of turbulence. Curious what other insights came out of #HAICON26? Both of our keynotes are now live on our YouTube channel: 🎤 Pierre Gentine (Columbia University): Lost (and found) in latent land: applications to weather and climate https://lnkd.in/ddg3yVvV 🎤 Cordelia Schmid (Google): Video-Guided Policies for Robotic Manipulation https://lnkd.in/dVgghGbR Our BAIOSPHERE session, invited speakers' talks, and all other sessions are coming to your feed soon - stay tuned! #HAICON26 #AI #ML

Video Content