Regularizing deep networks with prior knowledge: A constraint-based approach

作者：

Highlights：

• Definition of a neuro-symbolic approach to inject prior knowledge into deep learning.

• Stating the importance of FOL prior knowledge to improve the performances of deep architectures.

• Large scale evaluation of the proposed method on image classification tasks.

摘要

•Definition of a neuro-symbolic approach to inject prior knowledge into deep learning.•Stating the importance of FOL prior knowledge to improve the performances of deep architectures.•Large scale evaluation of the proposed method on image classification tasks.

论文关键词：Deep learning,Convolutional neural networks,Image classification,Neuro symbolic methods,First-order logic,Learning from constraints

论文评审过程：Received 2 December 2019, Revised 21 March 2021, Accepted 24 March 2021, Available online 2 April 2021, Version of Record 16 April 2021.

论文官网地址：https://doi.org/10.1016/j.knosys.2021.106989