BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:bec3387a634392e5ebeec91197e6405b
DTSTAMP:20260716T025716Z
SUMMARY:MFC CDT Summer School Talk – Serge Guillas (UCL) & Adam Sykulski 
 (Imperial)
DESCRIPTION:Serge Guillas\, UCL\nTitle- Novel climate modelling merging Mac
 hine Learning and High Performance Computing\nThis talk covers new advance
 s in the design\, development and deployment of Machine Learning (ML) tool
 s for climate modelling. We first present how using ML can improve convect
 ion within a climate model. The climate model’s temperature and humidity
  profiles are stochastically perturbed every 6 hours at runtime based on G
 aussian process emulators trained on data from a high-resolution model wit
 h realistic convection. This hybrid approach on a simplified climate model
 ’s output (SPEEDY) improves the precipitation pattern globally\, with th
 e largest reductions in the tropics by around 20%. A follow-on project exa
 mines the engineering challenges of compiling the Gaussian process deploym
 ent into a Fortran-based state-of-the-art climate model (CESM) at runtime 
 with minimal overhead. Running on an NVIDIA GH200 device reduces the overh
 eads from 3x to only 3% in compute time and yields large modelling improve
 ments. We then discuss the first steps towards including coastal wave mode
 lling into a climate model through ML using this hybrid strategy\, emulati
 ng Smoothed Particle Hydrodynamics modelling of wave breaking for air-sea 
 exchanges representations. We then present ongoing work on the modelling o
 f weather and climate with diffusion models to reflect better uncertaintie
 s\, with a focus on acceleration strategies of distillation\, and impact o
 n downstream risk quantification. Finally we introduce our current investi
 gation on the modelling and a possible early detection of tipping points i
 n the North Atlantic Subpolar Gyre\, requiring decadal multiphysics ML emu
 lation to enable the assessment of uncertainties for robust warnings.\nAda
 m Sykulski\, Imperial College London\nTitle- Transforming covariates to e
 nhance spatio-temporal predictions in climate applications\nIn multivariat
 e spatio-temporal statistics our starting point is often a linear model li
 ke multiple linear regression or vector autoregression. However\, sometime
 s the cross-sectional interactions between variables are somewhat more sub
 tle\, and exist only in the tails of the distribution\, or in some other n
 onlinear sense. In this talk we provide three practical case studies where
  transforming covariates provides improved models and predictions. The fir
 st looks at the nonlinear effects of extreme precipitation on deforestatio
 n in Nepal\, using novel approaches from functional data analysis. The sec
 ond looks at the nonlinear effects of ocean currents on the abundance of A
 ntarctic krill\, using novel approaches from spectral analysis. Finally\, 
 the third introduces a new methodology for performing causal discovery in 
 a nonlinear multivariate time series setting\, using novel approaches from
  extreme value analysis.\nSchedule\n\n\n\n\n\n Time\n\n\n Friday\, 5 Jun
 e 2026 (Hybrid via zoom and in Huxley 340)\n\n\n\n\n14.00 – 15:00\n\n\nS
 erge Guillas\n\n\n\n\n15:00 – 15:15\n\n\nBreak\n\n\n\n\n15:15 – 16:15\
 n\n\nAdam Sykulski\n\n\n\n\n
URL:https://www.imperial.ac.uk/events/210213/mfc-cdt-summer-school-talk-ser
 ge-guillas-ucl-adam-sykulski-imperial/
DTSTART;TZID=Europe/London:20260605T140000
DTEND;TZID=Europe/London:20260605T163000
LOCATION:340\, Huxley Building\, South Kensington Campus\, Imperial College
  London\, London\, SW7 2AZ\, United Kingdom
END:VEVENT
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
DTSTART:20260605T140000
TZNAME:BST
TZOFFSETTO:+0100
TZOFFSETFROM:+0100
END:DAYLIGHT
END:VTIMEZONE
END:VCALENDAR