Thesis Director: Prof. S. Kashyap,
\r\nComputer and Communication Sciences doctoral program
\r\nThesis Nr. 11376
\r\n
\r\nTo take part in the public defense, please contact directly the speaker
Poster Exhibition of Materials Science and Engineering Graduate Students, featuring their 2025–2026 Master’s Projects, in the MXF Entrance (MXF-194.0) from October 1st to 21st, 2026.
","image_description":"","creation_date":"2026-09-29T10:27:30","last_modification_date":"2026-10-05T10:56:23","link_label":"","link_url":"","canceled":"False","cancel_reason":"","place_and_room":"MXF 194.0","url_place_and_room":"https://plan.epfl.ch/?room==MXF%20194.0","url_online_room":"","spoken_languages":["https://memento.epfl.ch/api/v1/spoken_languages/1/?format=json","https://memento.epfl.ch/api/v1/spoken_languages/2/?format=json"],"speaker":"","organizer":"Materials Science and Engineering Section","contact":"Chrystelle Demierre","is_internal":"False","theme":"","vulgarization":{"id":1,"fr_label":"Tout public","en_label":"General public"},"registration":{"id":3,"fr_label":"Entrée libre","en_label":"Free"},"keywords":"Materials Science and Engineering Section Science et Génie des matériaux Posters","file":null,"icalendar_url":"https://memento.epfl.ch/event/export/121985/","category":{"id":5,"code":"EXPO","fr_label":"Expositions","en_label":"Exhibitions","activated":true},"academic_calendar_category":null,"domains":[],"mementos":["https://memento.epfl.ch/api/v1/mementos/8/?format=json"]},{"id":72764,"title":"IMX Talks - Resolving Hierarchical Structure with X-ray scattering imaging","slug":"imx-talks-resolving-hierarchical-structure-with-x","event_url":"https://memento.epfl.ch/event/imx-talks-resolving-hierarchical-structure-with-x","visual_url":"https://memento.epfl.ch/image/33978/200x112.jpg","visual_large_url":"https://memento.epfl.ch/image/33978/720x405.jpg","visual_maxsize_url":"https://memento.epfl.ch/image/33978/max-size.jpg","lang":"en","start_date":"2026-11-02","end_date":"2026-11-02","start_time":"16:30:00","end_time":"17:30:00","description":"The function of many materials, from biological tissues to industrial polymers, depends on how their nanoscale building blocks are arranged and aligned within a macroscopic structure, which is essential to understand but difficult to resolve: established high-resolution techniques typically access only small sample volumes, limiting the ability to capture structural or biological variability across extended, heterogeneous samples.
\r\nSmall-angle X-ray scattering (SAXS) imaging probes local nanostructure by raster-scanning a focused synchrotron beam across a sample. This can be extended to resolve nanostructure orientation voxel-by-voxel in three dimensions in a method called SAXS tensor tomography, which we implement in our open-source software package Mumott. In this presentation, I will show how this approach, and its continued development, can be applied across different material systems.
\r\nFor injection-molded polymer packaging, SAXS imaging has resolved the layered anisotropic nanostructure important for its mechanical properties, and, combined with industrial processing simulations, the structure formation during thermal processing. Another example is bone, where resolving structure requires both three-dimensional information and sufficient sampling to capture biological variability. Combining 2D and 3D SAXS imaging, we studied regional differences in the human femoral neck, an anatomical site relevant to hip fractures, and, at higher resolution, resolved alternating chirality between neighboring lamellae within individual osteons, a structural feature proposed on mechanical grounds over a century ago.
\r\nThese examples illustrate how tensor tomography, extended beyond X-ray scattering to wide-angle diffraction and visible-light polarimetry, provides a route to resolving hierarchical structure across a wide range of material systems.
\r\n
\r\nBio: Marianne Liebi studied food science at ETH Zurich, where she also completed her PhD in 2013, followed by postdoctoral research at the Paul Scherrer Institute and MAX IV Laboratory, Lund. In 2017, she started her own research group at Chalmers University of Technology, Gothenburg, and was appointed in 2021 as tenure-track Assistant Professor at EPFL, where she heads the Laboratory for X-ray Characterization of Materials (cam-X) within the Institute of Materials, in a joint position with the Paul Scherrer Institute (PSI). Her research develops advanced X-ray and optical imaging techniques, centered on small-angle X-ray scattering (SAXS) in 2D and 3D, to resolve the hierarchical structure of materials ranging from fiber composites to pharmaceutical formulations and bone.
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.
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