BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Walrus: A Cross-Domain Foundation Model for Continuum Dynamics - P
 ayel Mukhopadhyay (University of Cambridge)
DTSTART:20260203T140000Z
DTEND:20260203T150000Z
UID:TALK241630@talks.cam.ac.uk
CONTACT:Sven Krippendorf
DESCRIPTION:Foundation models have transformed machine learning for langua
 ge and vision\, but achieving comparable impact in physical simulation rem
 ains a challenge. Data heterogeneity and unstable long-term dynamics inhib
 it learning from sufficiently diverse dynamics\, while varying resolutions
  and dimensionalities challenge efficient training on modern hardware. Thr
 ough empirical and theoretical analysis\, we incorporate new approaches to
  mitigate these obstacles\, including a harmonic-analysis-based stabilizat
 ion method\, load-balanced distributed 2D and 3D training strategies\, and
  compute-adaptive tokenization. Using these tools\, we develop Walrus\, a 
 transformer-based foundation model developed primarily for fluid-like cont
 inuum dynamics. Walrus is pretrained on nineteen diverse scenarios spannin
 g astrophysics\, geoscience\, rheology\, plasma physics\, acoustics\, and 
 classical fluids. Experiments show that Walrus outperforms prior foundatio
 n models on both short and long term prediction horizons on downstream tas
 ks and across the breadth of pretraining data\, while ablation studies con
 firm the value of our contributions to forecast stability\, training throu
 ghput\, and transfer performance over conventional approaches.
LOCATION:DAMTP\, room MR4
END:VEVENT
END:VCALENDAR