解耦表征学习综述

在大数据时代下, 以高效自主隐式特征提取能力闻名的深度学习引发了新一代人工智能的热潮, 然而其背后黑箱不可解释的“捷径学习”现象成为制约其进一步发展的关键性瓶颈问题. 解耦表征学习通过探索大数据内部蕴含的物理机制和逻辑关系复杂性, 从数据生成的角度解耦数据内部多层次、多尺度的潜在生成因子, 促使深度网络模型学会像人类一样对数据进行自主智能感知, 逐渐成为新一代基于复杂性的可解释深度学习领域内重要研究方向, 具有重大的理论意义和应用价值. 本文系统地综述了解耦表征学习的研究进展, 对当前解耦表征学习中的关键技术及典型方法进行了分类阐述, 分析并汇总了现有各类算法的适用场景并对此进行了可视化实验性能展示, 最后指明了解耦表征学习今后的发展趋势以及未来值得研究的方向.

A Review of Disentangled Representation Learning

In the era of big data,deep learning has triggered the current rise of artificial intelligence which is known for its ability of efficient autonomous implicit feature extraction.However,the unexplainable“shortcut learning”phenomenon behind it has become a key bottleneck restricting its further development.By exploring the complexity of physical mechanism and logical relationship contained in big data,the disentangled representation learning aims to explore the multi-level and multi-scale explanatory generative latent factors behind the data,and prompts the deep neural network model to learn the ability of intelligent human perception.It has gradually become an important research direction in the field of deep learning,with huge theoretical significance and application value.This article systematically reviews the research of disentangled representation learning,classifies and elaborates state-ofthe-art algorithms in disentangled representation learning,summarizes the applications of the existing algorithms and compares the performance of existing algorithms through experiments.Finally,the challenges and research trends in the field of disentangled representation learning are discussed.