Sacha Braun

PhD student in machine learning with Francis Bach &
Michael I. Jordan: uncertainty quantification, conformal prediction, learning theory.

[Sacha Braun @PotoSacho]

I will be in Palermo for the Italian Meeting on Probability and Mathematical Statistics (IMPMS) next week if you want to chat ! (and present during the conformal prediction session on Tuesday)

[Sacha Braun @PotoSacho]

Want to directly minimize the conditional quantile of your predictions?🔥

With Michael I. Jordan & @BachFrancis we introduce SLS regression. We use volume as a direct optimization criterion for estimating conditional HDRs.

Plus: it’s conformal-ready.
📝https://t.co/hdgIRdbG53

[Sacha Braun @PotoSacho]

https://t.co/cyxmzIk9y6

[Sacha Braun @PotoSacho]

Amazing calibration workshop happening at AISTAT ! We will present a poster this afternoon (we propose a variational estimator for Lp calibration error - including L1!) you should definitely come if you still think you can reliably evaluate calibration errors with ECE :-)

[Eugène Berta @Eugene_Berta]

Check out the great list of papers accepted at our workshop on probability calibration, in particular the student and non-student outstanding paper awards!
See you in Tangier, looking forward to the great conversations about calibration and uncertainty in modern AI 👇

[Sacha Braun @PotoSacho]

@tylerfarghly @Sierra_ML_Lab @inria_paris @ENS_ULM @umutsimsekli Welcome !

[NeurIPS Europe Conference @EurIPSConf]

Today NeurIPS is announcing our official satellite event in Paris.

After responding to the call from Ellis following the success of EurIPS in December, we are pleased to reach a new milestone by joining forces with the NeurIPS organizing committee for the 2026 edition!

[Sacha Braun @PotoSacho]

It's worth a read!

[Eugène Berta @Eugene_Berta]

Excited to share that "Structured Matrix Scaling for Multi-Class Calibration" was accepted to #AISTATS2026!

The final version of the paper is out, and the updated implementation in our package probmetrics (version 1.1) is both FASTER & BETTER.

[Sacha Braun @PotoSacho]

@BasileTerv987 @JimmyTYYang1 @AdrienBardes @ylecun Congrats !

[Sacha Braun @PotoSacho]

📖 Read the full paper: https://t.co/7zKXLntspg
💻 Package available here: https://t.co/lmBYZScxBs

Use our package as :
pip install covmetrics
from covmetrics import ERT
print( ERT().evaluate(x, cover, alpha) )

[Sacha Braun @PotoSacho]

📢 How can you measure conditional miscoverage ??

With @DHolzmueller, Michael I. Jordan & @BachFrancis, we recast conditional coverage as classification, offering a new lens on both its meaning and its measurement. (+ release a package to compute our metric in 2 lines of code)!

[Sacha Braun @PotoSacho]

@Gauthier_E_ @Googleorg @BachFrancis @Inria @ENS_ULM Impressive !

[Francis Bach @BachFrancis]

Not all scaling laws are nice power laws. This month’s blog post: Zipf’s law in next-token prediction and why Adam (ok, sign descent) scales better to large vocab sizes than gradient descent: https://t.co/uoy5GPrZek

[Sacha Braun @PotoSacho]

Coming back from an amazing week in Warwick, discovering sampling and diffusion !

Big thanks to Alain Durmus for the teaching and the organization team (Hugo Queniat & Filippo Pagani) for the organization!

[Sacha Braun @PotoSacho]

@dtiapkin Congrats !

[Sacha Braun @PotoSacho]

@Eugene_Berta @BachFrancis Paper : https://t.co/rZksRiCql2
Code : https://t.co/PyVtk7uPax

[Sacha Braun @PotoSacho]

📢 New paper on arXiv alert!

With @Eugene_Berta, Michael I. Jordan & @BachFrancis we improve conditional coverage in conformal prediction + handle missing outputs + other extensions you should definitely check in the paper !🎯 #conformalPrediction

[Sacha Braun @PotoSacho]

I guess running in the mountains is a typical holiday break for researchers — and if you're not a runner, consider volunteering at a race instead and soak in all the good vibes! Thanks to the ut4m team for the organisation ;)

[Fabian Schaipp @FSchaipp]

Best tutorials are those that do not only promote the speakers' own work.

#ICML2025

[Francis Bach @BachFrancis]

What if AI isn’t about building solo geniuses, but designing social systems?
Michael Jordan advocates blending ML, economics, and uncertainty management to prioritize social welfare over mere prediction.
A must-read rethink.
https://t.co/HUJh97pq5N