arXiv:2006.06462cs.LG
Learning advanced mathematical computations from examples
Early discovery. Linked by one identifiable commenter. This paper is outside the main ranking; a mention is not an endorsement.
Abstract
Using transformers over large generated datasets, we train models to learn mathematical properties of differential systems, such as local stability, behavior at infinity and controllability. We achieve near perfect prediction of qualitative characteristics, and good approximations of numerical features of the system. This demonstrates that neural networks can learn to perform complex computations, grounded in advanced theory, from examples, without built-in mathematical knowledge.
They "can't reason" but they can solve symbolic integration and differential equations (https://arxiv.org/pdf/2006.06462.pdf), beat us at all board games, and more recently get to be almost impossible to tell apart from human writing (GPT3).
visargaView on HNParent context