Kokoyi: Executable LaTeX for End-to-end Deep Learning

Minjie Wang, Haoming Lu, Yu Gai, Lesheng Jin, Zihao Ye, Zheng Zhang

Abstract: Despite substantial efforts from the deep learning system community to relieve researchers and practitioners from the burden of implementing models with ever-growing complexity, a considerable lingual gap remains between developing models in the language of mathematics and implementing them in the languages of computer. The mission of Kokoyi is to close this gap by enabling automatic translation of mathematics into efficient implementations, thereby making math-in-codes and math-in-model consistent. This paper presents our first step towards the goal: kokoyi-lang, a programming language with the syntax of LaTeX and the semantics of deep learning mathematics, and a prototype kokoyi-lang compiler and runtime supporting advanced optimizations such as auto-batching. Kokoyi is integrated with Jupyter Notebook, and will be released in open-source.