Biosketch
Sarah has been thinking a lot about randomized algorithms and stochastic processes. She suspects that making progress in machine learning will require us to embrace its randomness, or at least get more comfortable with it. Tentatively the destination is interpretable models running on low-power edge devices for medical diagnostics and/or assistive technologies. She is always happy to hear about new ideas on how to get there.
Sarah is a PhD student at the Chair of Mathematical Foundations of Artificial Intelligence, advised by Prof. Dr. Gitta Kutyniok. Before this, she worked as a research assistant at NYU Abu Dhabi. Before that, she did an MS in Computer Science at NYU Tandon School of Engineering. Before that, she studied Studio Art at NYU Steinhardt. She still likes working in pastels, writing, ceramics, and photography, but her primary medium is now equations.
relAI Research
On the Hegemony of GEMM: Deep Learning as Numerical Method
My research views deep learning as the latest chapter in the history of numerical methods. This perspective can help us to understand models by applying knowledge and techniques developed for high performance and scientific computing. Ultimately, this connection can suggest how to make models more reliable, which will improve their usefulness for applications in science and engineering.