🚨 Want to join us? - We offer 5 postdoc positions in European projects in València - ML/DL, #Earth science, #hybrid modelling, #climate model #parametrization, #UQ, #diffusion & #foundation models - Apply by Aug 1st 2024: https://t.co/hnAtT8q4L7 Happy summer! 🌊🌞🥘⚽️

🚨New preprint🎉 "Generative Data Assimilation of Sparse Weather Station Observations at Kilometer Scales" https://t.co/FbfxNyeDvA An absolute pleasure working on this during my internship @nvidia with @NoahBrenowitz @SciPritchard @JaideepPathak @tropmetpie and others 🧵

📢Excited to share I've successfully defended my thesis at @oxcsml and will be joining @eapsMIT this summer as a postdoctoral associate🌍 Many thanks to my examiners @nicholls_geoff and Tom Beucler, to my supervisor @sejDino, and to @DWatsonParris for making this possible!

🌟 It’s a great honour that our work has been selected as a spotlight paper for #ICML2024 ! This work offers a new perspective on Domain Generalisation so check it out! Many thanks to collaborators @_anurags14 @shbouabid @krikamol!

I want to thank my wonderful co-authors @DWatsonParris @sejDino @iMIRACLI_ITN, but also @maybritt_sch and three anonymous reviewers for their great quality feedback on this work!

📢🥳FaIRGP is out! It's a mathematically tractable and easy-to-use probabilistic ML emulator of surface temperatures that uses FaIR as its backbone. This hybrid model improves over purely physics or data-driven emulators with uncertainty quantification! https://t.co/Go7nb99FYa

🚨 Our #ICML2024 paper contends that domain generalisation encompasses both statistical learning and decision-making. Learners are thus compelled to make normative judgments, leading to misalignment amidst institutional separation from model operators. @icmlconf

⚠️New paper out⚠️ Excited to share the publication of our latest work on the response of clouds to shipping emissions🚢 Satellite retrievals + modelled shipping pollution = insights into aerosol-cloud-climate change interactions.

On the academic job market this year? @CISPA offers an exciting opportunity to conduct world-class research with generous research support. 🚨 Tenure-Track Faculty in Artificial Intelligence and Machine Learning (f/m/d) RT Please 🙏 https://t.co/DIkoBtQXFf

Its hard to overstate just how exceptionally high global temperatures are at the moment. They have blown past anything we've previously experienced by a huge margin. Over at The Climate Brink, we try and visualize this summer of extremes in seven charts. https://t.co/yApwMbyxgG

Il y a 48h, une étude scientifique publiée annonçait que la 6ème limite planétaire sur 9 était désormais officiellement dépassée... et ce fut un vrai raz de marée médiatique ! Non bien sûr, comme d'habitude, tout le monde s'en fout.

📢Post-doc opening in Adelaide with Prof. Dino Sejdinovic @sejDino 📢 I highly recommend working with Dino, who was one of my supervisors during my PhD in Oxford :) Consider applying if you want to do a great post-doc in sunny Australia ! https://t.co/oV6M0DzcP4

By gaining trust in such a data-driven yet physically grounded model, we hope the climate science community can benefit more widely from their potential.

Finally, FaIRGP is easy-to-use (analytical expressions throughout), can naturally account for climate internal variability and provides principled uncertainty quantification through its Bayesian treatment.

One other cool feature is that, whilst this is not its primary objective, FaIRGP can also provide probabilistic estimates of the radiative forcing informed by temperature observations.

... and (3) can be extended to emulate spatial response, below the prediction bias for surface temperature anomalies. We also show that FaIRGP better takes into account the response to anthropogenic aerosol emissions, which can make a big difference.

... (2) also has a robust physical grounding which safeguards against the pitfalls of purely data-driven methods...

The result is an emulator for surface temperature anomaly that enjoys the flexibility of GPs and can (1) learn from observations how to deviate from FaIR but...

FaIRGP combines a physics-driven emulator (FaIR) with a data-driven model (Gaussian processes or GPs). The idea is fairly simple : take the FaIR temperature response model, but treat the radiative forcing as a GP.

Both categories have their strengths and weaknesses, but luckily the strengths of physics-driven emulators can compensate for the weaknesses of data-driven emulators, and vice-versa. This is why we propose FaIRGP, a hybrid physics/data-driven emulator.