一句话:提出在现有编程语言中引入新特性,使开发者能直接利用现代强化学习算法与实现。
强化学习在工业自动化、游戏等诸多领域取得了巨大成功,但普通开发者要将其能力引入自己的程序仍存在语言层面的门槛。本文提出「学习增强语言」这一方向:不必发明全新的语言,而是在现有编程语言中增添新的语言特性,使开发者可以方便地调用现代强化学习算法及其实现,从而把 RL 作为语言一等能力来编程。论文属于立场与愿景性工作,重点在于阐述这一设计方向的动机与可能形态,说明语言应当如何为学习型组件提供支持,而摘要中并未给出具体的形式化定义、实现方案或实证评估结果。
Reinforcement learning (RL) has seen tremendous success at solving a variety of problems ranging from industrial automation to games. This paper describes how existing programming languages can be augmented with new features so as to allow developers to exploit the power of modern RL algorithms and implementations.
DOI 原文 ·
@inproceedings{paperbot4668,
title = {Towards learning-augmented languages},
author = {Xinyuan Sun},
booktitle = {FSE},
year = {2018},
doi = {10.1145/3236024.3275432}
}