Aurelien Lucchi

Researcher in optimization and theoretical machine learning. Associate professor at the University of Basel. Past: EPFL, ETH Zurich (Switzerland) 🇨🇭

[Aurelien Lucchi @AurelienLucchi]

@thsottiaux Codex does not follow instructions. E.g. it does not work well with autoresearch (it stops all the time even when tell it not to). Claude Code is much better at this task. Hope you guys catch up.

[Ilija Bogunovic @ilijabogunovic]

Hiring 2 summer ML research interns at the University of Basel 🇨🇭.

Research topics: RL/diffusion LLM post-training, reasoning, or LLM orchestration. Possible fully funded PhD offers to follow.

I'll be at ICLR this week and happy to chat.
Apply: https://t.co/gOdLRv771L

[Aurelien Lucchi @AurelienLucchi]

After a lot of hands-on time with Codex and Claude: Claude is significantly better at actually following instructions. Not even close right now.

[Tibo @thsottiaux]

Codex

Compute efficient ✅
Always up, never down ✅
Best at hardcore engineering ✅
Crazy good app, first to escape the terminal ✅

[Aurelien Lucchi @AurelienLucchi]

Why Basel / Switzerland (lifestyle + funding matters)

📍 Basel = top research environment + high quality of life
💰 fully funded PhD with competitive Swiss salary

[Aurelien Lucchi @AurelienLucchi]

🚨 PhD position in Reasoning for LLMs at the University of Basel 🇨🇭

Work on:
• reasoning in LLMs
• diffusion LLMs
• theory ↔ real-world applications

Top venues (ICML, NeurIPS, ICLR) + strong math/ML focus

Joint position with I. Bogunovic @ilijabogunovic

[Aurelien Lucchi @AurelienLucchi]

Apply👇
https://t.co/5rI0FmUTeD

[Rustem @ruuustem_10]

Excited to share our latest work on bridging theory & practice in optimization 🚀

We study stochastic conditional methods with momentum and provide practical strategies for choosing batch size and Frank–Wolfe stepsizes when token budget increases

Paper: https://t.co/ckinOF40Zw

[Juliusz Ziomek @JuliuszZiomek]

🚨 New Benchmark Alert!! 🚨
Navigate Wikipedia hyperlinks step-by-step.
No map.
Just planning and world knowledge!

We evaluated 20+ models on 3 difficulty levels:
Gemini-3: 95% → 66% → 23%
GPT-5: 92.5% → 60% → 15%
Opus 4.5: 91.5% → 56% → 18%
We discover a Planning Gap!🧵

[Aurelien Lucchi @AurelienLucchi]

@ilijabogunovic 100%. I don't think this solves the problem and it introduces more randomness in the process as we will have new ACs who are unfamiliar with the papers and outdated reviews/scores.

[Ilija Bogunovic @ilijabogunovic]

Want to do a PhD on RL for LLMs? 🇨🇭🤖🎓
One week left to apply!

[Aurelien Lucchi @AurelienLucchi]

And Switzerland offers a remarkable diversity of landscapes, from alpine peaks to tranquil lakes.

[Aurelien Lucchi @AurelienLucchi]

The University of Basel has an opening for a Professorship in Deep Learning. We are growing, join us!
https://t.co/ayIERM46R1

[Aurelien Lucchi @AurelienLucchi]

The conditions to create a group in Switzerland are among the best in the world: significant fundings for PhD students, high acceptance rate from the funding agency SNF, access to European funding opportunities, lots of amazing PhD candidates, great research partners, etc.

[Aurelien Lucchi @AurelienLucchi]

@ilijabogunovic and I are looking for two PhD candidates in the field of reasoning in machine learning. Apply here: https://t.co/R7UabXj0Yr

[Aurelien Lucchi @AurelienLucchi]

Post-doc position in the field of Optimization and Deep Learning Theory
https://t.co/QcOZXezmHn

PhD position in the field of Optimization and Deep Learning Theory
https://t.co/Yw5J4PSZmL

PhD position in the field of Reasoning in Machine Learning
https://t.co/GzCJG8Tn0E

[Aurelien Lucchi @AurelienLucchi]

Our research group in the department of Mathematics and CS at the University of Basel (Switzerland) is looking for several PhD candidates and one post-doc who have a theoretical background in optimization and machine learning or practical experience in reasoning. RT please.

[Aurelien Lucchi @AurelienLucchi]

We have a non-vanishing term in the convergence rate, and interestingly this term decreases when we increase the number of parameters (confirmed experimentally and in line with many other prior works in the field):

[Aurelien Lucchi @AurelienLucchi]

Loss Landscape Characterization of Neural Networks without Over-Parametrization to appear at @NeurIPSConf 2024.

We propose a novel class of functions that characterize the loss landscape of deep models without requiring over-parametrization.

https://t.co/9ydIXDkQJf

[Aurelien Lucchi @AurelienLucchi]

We also characterize the convergence of SGD (and other optimizers):