BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Memento EPFL// BEGIN:VEVENT SUMMARY:IC Colloquium: Imaging at the Edge of Science: Integrating Scienti fic Knowledge and AI to Recover Hidden Structure DTSTART:20260202T101500 DTEND:20260202T111500 DTSTAMP:20260427T133440Z UID:038078683fe39447b575a1cbb5ac9236eeab05e4efaf6351452c25e8 CATEGORIES:Conferences - Seminars DESCRIPTION:By: Berthy Feng - MIT\nFaculty candidate\n\nAbstract\nImages p lay a central role in scientific discovery. Whether it’s astronomical\, biological\, or materials systems\, bringing complex phenomena into view e nables scientists to probe\, model\, and fundamentally understand them. Ho wever\, many of the most important scientific questions lie at the edge of what can be directly observed.\nWe can accomplish extreme imaging through computational methods\, bringing the invisible into view by supplementing limited observable data with human-imposed assumptions\, or priors. When imaging for science\, the challenge is imposing just enough known assumpti ons to infer the unknown.\nI create principled methods for bringing advanc ed priors\, such as scientific knowledge and AI\, into computational imagi ng. Using astrophysics as a running example\, this talk presents my vision for a framework in which scientists systematically explore different prio rs\, understand their effects on imaging\, and extract scientific insights .\nThe talk is organized in three parts.\n\n First\, we understand the imp ortance of priors in extreme scientific imaging. I present my work on leve raging generative AI to flexibly tune a knob between different priors and understand their effects on imaging. Applied to black-hole imaging\, my ap proach lets us infer physical features of a real black hole by identifying image features that are robust to prior assumptions.\n Second\, we carefu lly balance scientific assumptions to solve an extreme imaging problem in astrophysics. I present PINeRF\, a method for imaging the dynamic 3D gas n ear a black hole. PINeRF strikes a balance between known/unknown physics\, imposing known physics as hard constraints on the solution while leaving room for learning unknown physics\, such as the velocity field near the bl ack hole.\n Third\, we open an efficient route for bringing in known physi cs across imaging problems. I present Neural Approximate Mirror Maps (NAMM s)\, which learn to automatically impose any desired physics constraint on to any image. With NAMMs\, we can easily incorporate known constraints (e. g.\, conservation laws) into generated and reconstructed images.\n\nThe id eas of my talk naturally extend to many scientific domains\, including bio logy\, chemistry\, and materials science.\n\nBio\nBerthy Feng is a postdoc toral researcher at MIT CSAIL and a fellow at the NSF Institute for AI and Fundamental Interactions (IAIFI)\, working with Prof. Bill Freeman. She r eceived her PhD in Computational and Mathematical Sciences at Caltech\, wo rking with Prof. Katie Bouman. Before that\, she received her Bachelor’s degree in Computer Science at Princeton University. She creates principle d methods for scientific imaging that leverage data-driven and scientific knowledge.\n\nMore information\n  LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420 https://epfl.zoom.us/ j/62670608023 STATUS:CONFIRMED END:VEVENT END:VCALENDAR