Moritz H

IMPRS-IS PhD student with @UlrikeLuxburg and @bedartha, Theory of Machine Learning, ML in Climate Science

[Moritz H @moritz_haa]

@2prime_PKU Nice work! You will probably interested in this one too:
Mind the spikes: Benign overfitting in kernels and neural networks in fixed dimension https://t.co/bISAnQL5tZ

[Moritz H @moritz_haa]

I will present our work on benign overfitting of kernels and neural networks at #NeurIPS2023 on Thursday 5pm (#1724).

Feel free to reach out if you want to chat about (deep) learning theory towards principled deep learning, also beyond kernels.

Find a thread below.

[David Holzmüller @ ICML 2026 @DHolzmueller]

Can kernel methods and wide neural networks overfit benignly in fixed dimension?

In our NeurIPS 2023 paper with @moritz_haa, @UlrikeLuxburg, and Ingo Steinwart, we prove: yes, but only with some non-standard modifications. 1/ 🧵

[Ulrike Luxburg @UlrikeLuxburg]

📷
Benign overfitting even works in small, fixed dimension! Not with ReLu activation function or smooth kernels, but with extra-spiky variants. #neurips2023 paper with @moritz_haa, @DHolzmueller, Ingo Steinwart:
https://t.co/Al0Wl5Hlwf

[David Holzmüller @ ICML 2026 @DHolzmueller]

Can kernel methods and wide neural networks overfit benignly in fixed dimension?

In our NeurIPS 2023 paper with @moritz_haa, @UlrikeLuxburg, and Ingo Steinwart, we prove: yes, but only with some non-standard modifications. 1/ 🧵

[Moritz H @moritz_haa]

Many thanks to the organizers! I learned a lot, gained many new perspectives and had a lot of fun along the way! An invaluable experience.

[Lenka Zdeborova @zdeborova]

The past two weeks were a blast in Cargèse at "Statistical Physics & Machine Learning Back Together Again" https://t.co/dCnVSN32Ku. Around 100 of the top people in the field, including the next generation, discussed a lot of great science. I will miss you guys. We will be back!

[Ulrike Luxburg @UlrikeLuxburg]

"Pitfalls of Climate Network Construction: A Statistical
Perspective". With @moritz_haa and @bedartha, we show that many observations in climate networks might be spurious artifacts.
And I am sure the same would hold for brain networks.

https://t.co/wJNlAh8BIg