Directeurs de thèse : Prof. R. Logé, Dr E. Cantergiani
\r\nProgramme doctoral en Manufacturing
\r\nThèse n° 11916
\r\n
\r\nPour participer à la soutenance publique, merci de contacter directement l’intervenant
Thesis Director: Prof. D. Constam,
\r\nMolecular Life Sciences doctoral program
\r\nThesis Nr. 11220
\r\n
\r\nTo take part in the public defense, please contact directly the speaker
This talk is meant to give an overview of the semantics of HoTT in (∞,1)-categories, tackling the key challenge of rigidification: the process of bridging homotopy-coherent notions phrased in models of higher categories such as quasicategories, and their type-theoretical counterparts that must account for the inherent strictness of type theory.
\r\nAfter quickly recalling the usual categorical semantics of extensional Martin-Löf type theory within locally cartesian closed category, the plan is to present some useful ideas to carry over some of the argument to the higher setting. Specifically, I will discuss the process of turning an elementary higher topos into a model of HoTT, and sketch important ideas used to prove the internal language conjecture for locally cartesian closed (∞,1)-categories.
\r\n
\r\n
Thesis Directors: Prof. L. Shchutska, Prof. E. Graverini
\r\nPhysics doctoral program
\r\nThesis Nr. 11987
\r\n
\r\nTo take part in the public defense, please contact directly the speaker
Thesis Director: Prof. J. Waser,
\r\nChemistry and Chemical Engineering doctoral program
\r\nThesis Nr. 12072
\r\n
\r\nTo take part in the public defense, please contact directly the speaker
The talk is jointly organized by the EPFL AI Center and the MLO Lab as part of the AI fundamentals seminar series.
\r\n
\r\nHosting Professor: Prof. Martin Jaggi
\r\n
\r\nTitle
\r\nSpeedrunning SWE-bench, or the $1,000 coding agent
\r\n
\r\nAbstract
\r\nSpeedrunning has become a popular engineering challenge in machine learning: achieve the best performance within a fixed compute budget. We use it as a scientific tool to study how models acquire specialized skills and how far those skills transfer. We speedrun SWE-bench by training coding agents from scratch on demonstrations from stronger models. With about $1,000 of training compute, we reach a 25% solve rate, outperforming GPT-4o. Compared to models with extensive pre-training, our speedrun models solve largely the same SWE-bench problems, and generalize about as well to new bugs and codebases. What, then, does pre-training buy for SWE-bench? Our experiments suggest a one-student hypothesis: pre-training helps a student imitate its teacher, but at matched imitation loss, pre-training choices barely matter. We stress-test this hypothesis by comparing a model trained on the web against a “vintage\" one trained only pre-1931 documents. Up to the SWE-bench scores we reach, pre-training is interchangeable with specialization data. We release nanoswe as an open project for studying specialization through speedrunning.
\r\n
\r\nBio
\r\nRicardo Olmedo is a final-year PhD student at the Max Planck Institute for Intelligent Systems in Tübingen and a Google PhD Fellow. Advised by Moritz Hardt and Bernhard Schölkopf, he works on language model benchmarking and specialization. His research has earned three oral presentations at ICLR and NeurIPS. Most recently, he spent spring and summer 2026 visiting Sanmi Koyejo at Stanford University.