Andreas Mueller

Hello (better) world!

The newest release of sklearn includes pandas dataframe output (with feature names!!!) You should definitely try it out!

https://scikit-learn.org/stable/auto_examples/release_highlights/plot_release_highlights_1_2_0.html

I'm currently looking for interns working on transformers for tabular data, building on the ideas of TabPFN. Either spring 2024 or summer 2024 are possible, in person in CA. You need to be enrolled in an MS or PhD program. https://jobs.careers.microsoft.com/global/en/job/1657017/Research-Intern---Gray-Systems-Lab-(GSL), reach out for more info.

In case you missed it, definitely check out the interview of Geoffrey Hinton on MIT Technology Review about why he stepped down from Google and what he thinks are risks of current models: https://www.technologyreview.com/2023/05/03/1072589/video-geoffrey-hinton-google-ai-risk-ethics/

There's still tickets available for PyData Seatltle April 26th-28th https://pydata.org/seattle2023/

Reading a blog post summarizing a paper, I thought to myself "well this is not structured very well... I wonder if it's GPT generated. But at most 3.5, 4 would be better" So I wonder if people will start evaluating human performance by equating it to versions of language models (cause I apparently already started if this wasn't machine generated).

Really in-depth comparison of GPT-4 capabilities from Microsoft Research: https://arxiv.org/pdf/2303.12712.pdf

Now if only GPT-4 had enough tokens to summarize a 140 page document.

Amazing talk with so many insights on Bayesian Optimization at Meta by Eytan Bakshy. Great stuff on learning preference learning for multiple objectives beyond pareto frontiers http://arxiv.org/abs/2303.15746 #AutoML23

This article written by ChatGPT pointing out challenges for the future of AI is gold:
https://www.linkedin.com/pulse/prompt-engineering-101-introduction-resources-amatriain/

My keynote about challenges in going from AutoML to Automatic Data Science from the #AutoML conference is now on youtube: https://www.youtube.com/watch?v=jp_UZoM_OjE&list=PLp7L30nGpKM95zqSpkpMW8_c6EM8AsK6-&index=5

I'm grateful to Frank Hutter and the program committee for inviting me to #AutoML23 to speak about challenges of AutoML in industry and how we can better serve the day-to-day challenges of data scientists. You can find the slides here:
https://github.com/amueller/talks_odt/blob/master/2023/From%20AutoML%20to%20AutoDS.pdf
(thanks to Alexander Tornede for the photo)