Student Learning Seminar on Mathematics and Computation
Organizers: Henry Bosch and Shurui Liu
Upcoming Events
Starting from first principles, I will derive (a variant of) the GRPO algorithm, one of the most widely used algorithms for post-training large language models. Then I will sketch how this algorithm is implemented at scale. Finally, I will briefly describe an important open problem known as…
TBA: Mathematical Questions in Pretraining
My first productive experience using large language models occurred in October 2024. At the time, we were thinking about the hot spots conjecture, and a very lucky prompt gave us a key insight that helped us establish an interesting result on the extremal failure of this conjecture. I will…
Past Events
In this talk, I will attempt to survey the recent applications of neural networks to PDEs with a focus on two approaches. I will start by introducing physics-informed neural networks and how they have been applied to discovering unstable singular solutions to fluid equations. Then, I will finish…
I will describe some potential (but thus far largely unsuccessful) applications of ML techniques to proof discovery in geometric analysis.
Machine learning is usually presented as function approximation: in supervised learning, one aims to recover an unknown map from inputs to outputs, and results such as universal approximation theorems and generalization bounds explain why neural networks can, in principle, learn rich function…
We will meet to discuss organizational matters (topic of interest, etc.)