BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:Keynote Speaker - Statistical Breakthroughs and Novel Perspectives in Deep Learning Theory - Sophie Langer (Ruhr-Universität Bochum) DTSTART:20250828T083000Z DTEND:20250828T093000Z UID:TALK234508@talks.cam.ac.uk DESCRIPTION:Since several years\, deep learning has emerged as a transform ative field\, with its theory involving several disciplines such as approx imation theory\, statistics and optimization. In this talk we delve into k ey theoretical breakthroughs\, with a particular focus on statistical resu lts. We critically question prevailing frameworks and identify their key l imitations. Central to the discussion is a novel statistical framework for image analysis that reinterprets images not as high-dimensional entities\ , but as structured objects shaped by geometric deformations such as trans lations\, rotations\, and scalings. Within this framework\, classification is reframed as the task of learning uninformative deformations\, leading to convergence rates with more favorable trade-offs between input dimensio n and sample size. This geometric-statistical perspective not only provide s new guarantees for approximation and convergence in deep learning-based image classification but also prompts a rethinking of theoretical approach es for broader prediction problems. In the final part of the talk\, we exa mine the expressive power of ReLU networks in comparison to networks with Heaviside activation functions. While ReLU-based models have become standa rd in deep learning theory\, Heaviside networks offer a compelling alterna tive that aligns more closely with biologically inspired architectures. We conclude by outlining future research directions and reflecting on the ro le of theory in the field.\n \;\nThis talk is based on joint work with Juntong Chen\, Insung Kong and Johannes Schmidt-Hieber LOCATION:Seminar Room 1\, Newton Institute END:VEVENT END:VCALENDAR