Proper ML model testing is essential for robust & fair deployment, but how do we evaluate models on small subgroups or in domains for which we have too little or no test data? Nabeel Seedat, Boris van Breugel, Fergus Imrie & I marry model evaluation & synthetic data together!

Are you at #NeurIPS23 and interested in Generative Models and Synthetic Data? Let's meet! Today at 10:45-12:45, I'll be presenting Zhaozhi Qian, Rob Davis, and Mihaela van der Schaar SynthCity github.com/vanderschaarla…, a super useful resource for synthetic data generation and evaluation🚀

How do we stress-test ML models for settings for which we have no or little real data? #SyntheticData generated by #GenerativeModels may be the answer. Join us at #NeurIPS2023, 5pm (#907) to hear more. w/ Nabeel Seedat*,Fergus Imrie,Mihaela van der Schaar Link: neurips.cc/virtual/2023/p…

Introducing RadEdit! We use diffusion image editing to stress-test biomedical vision models: diagnosing failures and helping avoid extra data collection. This is my first paper since joining Microsoft Research and started as an internship project with Pedro P. Sanchez & Boris van Breugel

In our #ICML2024 position piece, we argue why tabular #foundationmodels deserve more research attention. I immensely enjoyed writing the piece, hopefully it's an interesting read. Stay tuned for more news in this area soon 🙊

Though unfortunately I couldn't be at ICML 2024, glad to see our position paper on the importance of tabular data and foundation models is getting some attention (no doubt thanks to the enthusiasm of Mihaela van der Schaar) 😊 arxiv.org/abs/2405.01147