Deep Learning for Symbolic Mathematics - Guillaume Lample & Francois Charton

Explore cutting-edge applications of deep learning in symbolic mathematics through this Stanford seminar presented by Guillaume Lample and Francois Charton. Delve into the fundamental concepts, starting with basic intuitions and progressing through advanced topics such as symbolic integration, ordinary differential equations, and dataset generation. Learn about the innovative model developed for tackling complex mathematical problems, and compare its performance against established tools like Mathematica. Examine the challenges and successes in generalization, and gain insights into equivalent solution representations. This 53-minute talk offers a comprehensive overview of the intersection between deep learning and symbolic mathematics, providing valuable references for further exploration in this exciting field.

Syllabus

Introduction.
Deep learning for symbolic mathematics.
Starting point.
Basic intuition.
The plan.
From expressions to trees.
Generating data.
Symbolic integration (forward approach).
Symbolic integration (backward approach).
Symbolic integration (integration by parts).
Ordinary Differential Equations (order 1).
Ordinary Differential Equations (ODE) - orde.
Ordinary Differential Equations (order 2).
Datasets.
The model.
Evaluation.
Comparison with Mathematica.
Integration-generalization issues.
Generalization - looking bad.
Generalization - looking better.
Generalization - looking forward.
Generalization - a fun fact.
Inside the beam - Equivalent solutions.
References.