Astrocytes are the predominant glial cells in the mammalian central nervous system (CNS) and play essential roles in maintaining neural circuit activity and tissue homeostasis. In response to CNS insult, including traumatic injury or implantation of a medical device, astrocytes undergo a conserved, temporally regulated reprogramming that reshapes their prioritized functions to protect adjacent neural tissue. The nature and extent of astrocyte reprogramming critically influences outcomes after CNS insult. For example, regeneration after traumatic CNS injury in adult mammals is limited by local responding astrocytes lacking sufficient self-renewal capacity. By contrast, the CNS of neonate (newborn) mammals possesses exceptional parenchymal repair capacity owing to immature astrocytes with elevated proliferative and migratory potential. In the context of an implanted medical device, the spatial extent of astrocyte reprogramming and how impacted astrocytes respond to any further device micromotion, latent infection, or other persistent device-derived stimuli, dictates device performance and longevity. Engineering astrocyte responses to promote targeted functional gains while preserving key capabilities of healthy neural tissue astrocytes will be critical to enabling effective neural regeneration therapies and improving long-term performance of implanted devices. In this talk, I will outline our emerging understanding of how adult astrocytes adaptively reprogram in response to CNS injury and implanted foreign bodies, drawing insights from innovative astrocyte-specific transcriptomic bioassays. I will also provide an overview of biomaterial and cell-grafting strategies that we are exploring to direct astrocyte responses in preclinical models of stroke, spinal cord injury, and foreign body responses.
\r\n
\r\nBio:
\r\nDr. Timothy O’Shea, PhD is an Assistant Professor in the Biomedical Engineering department at Boston University (BU). Tim grew up in Brisbane, Australia and completed a Bachelor of Medical Engineering (First Class Honours) and a Masters in Engineering Management at Queensland University of Technology (QUT) before moving to the US for graduate school. He completed his PhD study in Medical Engineering and Medical Physics within the collaborative Health Sciences and Technology program of the Harvard Medical School and the Massachusetts Institute of Technology where we conducted PhD thesis research with Institute Professor Robert Langer. As a Postdoctoral Fellow in Neurobiology at UCLA, he worked with Professors Michael Sofroniew (Neurobiology) and Timothy Deming (Bioengineering) developing and testing novel bioengineering tools to study biological mechanisms involved in CNS injury and repair as well as the CNS foreign body response to implants. Since Fall 2020 he has been leading the Glia Engineering Lab at BU where he has received numerous awards including the Maximizing Investigators' Research Award (MIRA) from NIH and grants from several spinal cord injury research foundations such as Craig H Neilsen, Paralyzed Veterans of America, Bryon Reisch, and Wings for Life.
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.
The Swiss Health Valley community comes together once a year to connect, exchange ideas and advance life sciences. In 2026, the BioAlps Networking Day will take place in Lausanne on 19 November at the Palais de Beaulieu.
\r\n
\r\nThe morning programme will showcase the strength and diversity of the Canton of Vaud’s life sciences innovation ecosystem. In the afternoon, discussions will focus on this year’s theme: Advancing Healthcare Through Digital Innovation.
\r\n
\r\nResearchers, entrepreneurs, investors and industry leaders are invited to explore emerging trends, build new partnerships and help shape the future of healthcare.\r\n
![\"\"](\"https://foster-lead-promote.ch/wp-content/uploads/2026/01/CG_EPFL_2026.png\")
\r\nIn all professions including academic careers, satisfaction and a sense of fulfillment are not always at hand, leading sometimes to questioning professional choices. Have you experienced not being completely sure of what is professionally right for you? Have you wondered how you could find it out? Do you feel you might be not well equipped to decide on the next step in your career, or to lack orientation? In this workshop, you will be provided with tools and orientation models that will guide you in your career decisions, and implement them. These are based on an authenticity-oriented perspective aiming at a fulfilling work and self-realization experience.
","image_description":"","creation_date":"2026-07-14T11:21:21","last_modification_date":"2026-07-14T11:23:17","link_label":"Website","link_url":"https://foster-lead-promote.ch/our_services/coachinggroups/","canceled":"False","cancel_reason":"","place_and_room":"EPFL","url_place_and_room":"","url_online_room":"","spoken_languages":["https://memento.epfl.ch/api/v1/spoken_languages/2/?format=json"],"speaker":"Afi Sika KuzeawuGenerative AI has become a part of students' everyday working life. If used without guidance, GenAI can cause over-reliance and leave students with a distorted sense of what they have actually learned. Most research and practice have focused on how we can scaffold students' GenAI use, for example through prompts, guardrails, or redesigned tasks. This talk asks a simpler question that has received far less attention. When should students be allowed to use AI so that it helps their learning?
\r\n
\r\nIn this lunch&LEARN session, Janne Rotter turns to a question that's been largely overlooked: not how, but when should students be allowed to use AI for it to actually support learning? Janne will present findings from a study of 105 university students comparing three conditions: unrestricted AI access, no access at all, and a \"timed\" condition where students could only turn to AI after first attempting the task on their own.
\r\n
\r\nDrawing on multiple pedagogical theories to build a decision rule for when access should open up, the results show that students in the timed condition outperformed those with unrestricted access on an objective post-test, and had a more accurate sense of their own understanding, without an explicit metacognitive prompting. Compared to withholding AI entirely, timed access also seemed to reduce task errors and time spent on task.
\r\n
\r\nJoin us to learn more about the topic and engage in further discussion.
\r\n