深度学习的范畴化基础:综合综述
机器学习
2024-10-16 v2 人工智能
范畴论
摘要
机器学习研究的空前进展速度带来了巨大成就,但也带来了严峻挑战。目前,该领域缺乏坚实的理论基础,许多重要成就源于非正式的设计选择,这些选择在原则上难以解释,且其有效性往往无法得到解释。研究债务不断增加,许多论文发现无法复现。本文是一篇综述,涵盖了最近尝试从范畴论角度研究机器学习的工作。范畴论是抽象数学的一个分支,已在众多领域成功应用,包括数学内部及外部。作为数学和科学的通用语言,范畴论可能为机器学习领域提供统一的结构,从而解决上述问题。本文主要关注范畴论应用于深度学习的应用。具体而言,我们讨论了利用范畴光学模型化基于梯度的学习、利用范畴代数和积分变换将经典计算机科学与神经网络联系起来、利用函子连接不同抽象层次并保持结构,最后讨论了利用字符串图为神经网络架构提供详细表征。
引用
@article{arxiv.2410.05353,
title = {Towards a Categorical Foundation of Deep Learning: A Survey},
author = {Francesco Riccardo Crescenzi},
journal= {arXiv preprint arXiv:2410.05353},
year = {2024}
}
备注
In the previous version of the survey, it was stated that the paper "Pooling Image Datasets with Multiple Covariate Shift and Imbalance" (Chytas, Lokhande, Singh) had been withdrawn by the authors. I have been informed that only an incomplete draft of the work was withdrawn after it was inadvertently uploaded. The complete work was actually published at ICLR and has never been withdrawn