BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:EESS talk on "Atmospheric Physics-Guided Machine Learning: Towards
  Physically-Consistent\, Data-Driven\, and Interpretable Models of Convect
 ion"
DTSTART:20220920T121500
DTEND:20220920T131500
DTSTAMP:20260509T025331Z
UID:ea639e67d9c19363929f50a5a36a84d7903ea2709f0b5d4d21b8d9d4
CATEGORIES:Conferences - Seminars
DESCRIPTION:Dr Tom Beucler\, Assistant professor\, Institute of Earth Surf
 ace Dynamics\, UNIL\nAbstract:\nData-driven algorithms\, in particular neu
 ral networks\, can emulate the effect of unresolved processes in coarse-re
 solution climate models if trained on high-resolution simulation data. How
 ever\, they may violate key physical constraints and make large errors whe
 n evaluated outside of their training set. I will share progress towards o
 vercoming these two challenges in the case of machine learning the effect 
 of subgrid-scale convection and clouds on the large-scale climate. First\,
  physical constraints can be enforced in neural networks\, either approxim
 ately by adapting the loss function or to within machine precision by adap
 ting the architecture. Second\, as these physical constraints are insuffic
 ient to guarantee generalizability\, I additionally propose to physically 
 rescale the inputs and outputs of machine learning algorithms to help them
  generalize to unseen climates. Overall\, these results suggest that expli
 citly incorporating physical knowledge into data-driven models of climate 
 processes may improve their consistency\, stability\, and ability to gener
 alize across climate regimes.\n\nShort biography:\nTom Beucler is an assis
 tant professor of environmental data science at the University of Lausanne
  in Switzerland. He recently started a lab specifically dedicated to the i
 ntersection of atmospheric physics and machine learning\, with the goal of
  improving our understanding of atmospheric dynamics and assisting weather
  and climate predictions. For that purpose\, his research group combines p
 hysical theory\, computational science\, statistics\, numerical simulation
 s\, and observational analyses. Before that\, Tom studied the interaction 
 of tropical storms\, radiation\, and atmospheric water as part of his PhD 
 at MIT. As a postdoc and project scientist at Columbia and UC Irvine\, he 
 investigated how to best integrate physical knowledge into neural-network 
 representations of convection for climate modeling\, which will be the the
 me of today’s webinar.\n 
LOCATION:GC B1 10 https://plan.epfl.ch/?room==GC%20B1%2010 https://epfl.zo
 om.us/j/63802032408
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR