Learning advanced mathematical computations from examples

Francois Charton, Amaury Hayat, Guillaume Lample

Keywords: differential equations, computation, transformers, deep learning

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.

One-sentence Summary: We train transformers to predict qualitative and numerical properties of differential equations

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Code: [ facebookresearch/MathsFromExamples](https://github.com/facebookresearch/MathsFromExamples)

Data: [Action Recognition in the Dark](https://paperswithcode.com/dataset/action-recognition-in-the-dark)