Interpretable Image Recognition Models for Big Data With Prototypes and Uncertainty

Author:

Wang Jingqi1ORCID

Affiliation:

1. Southwest University, China

Abstract

Deep neural networks have been applied to big data scenarios and have achieved good results. However, trust in high-accuracy deep neural networks is not necessarily achieved when facing high-risk decision scenarios. Every wrong decision has incalculable consequences in medicine, autonomous driving, and other fields. For an image recognition model based on big data to gain trust, it is necessary to simultaneously solve the interpretability of decisions and risk predictability. This paper introduces uncertainty into a self-explanatory image recognition model to show that the model can interpret decisions and predict risk. This approch enables the model to trust its decisions and explanations and to provide early warning and detailed analysis of risky decisions. In addition, this paper introduces the process of using uncertainty and explanations to achieve model optimization, which will significantly improve the application value of the model in high-risk scenarios. The scheme proposed in this paper solves the trust crisis caused by using black box image recognition models.

Publisher

IGI Global

Subject

General Computer Science

Reference24 articles.

1. Bayesian SegNet: Model Uncertainty in Deep Convolutional Encoder-Decoder Architectures for Scene Understanding

2. A case-based interpretable deep learning model for classification of mass lesions in digital mammography

3. Chai, L. R. (2018). Uncertainty estimation in Bayesian neural networks and links to interpretability [Master's Thesis]. Massachusetts Institute of Technology.

4. Data Uncertainty Learning in Face Recognition

5. This Looks like That: Deep Learning for Interpretable Image Recognition.;C.Chen;Proceedings of the 33rd International Conference on Neural Information Processing Systems,2019

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3