Trustworthy AI and Data Lineage

IF 3.7 4区计算机科学 Q1 COMPUTER SCIENCE, SOFTWARE ENGINEERING

IEEE Internet Computing Pub Date : 2023-11-17 DOI:10.1109/mic.2023.3326637

Elisa Bertino, Suparna Bhattacharya, Elena Ferrari, Dejan Milojicic

引用次数: 0

Abstract

AI trustworthiness properties are at the top of concerns for industry, governments, and academia. However, the AI and its models are only as good as the data used to train it. Data lineage could be tracked in many ways, including using metadata, from its generation usage, deployment, and verification. New standards, blueprints, best practices, and repositories for data are required to address requirements for data trustworthiness, such as sustainability, scale, and responsiveness but also ethics, diversity, equity, and inclusion. In this special issue of IEEE Internet Computing, we feature three articles. The first one addresses certification for trustworthy machine-learning-based applications, the second one is on the topic of data and configuration variances in deep learning, and the third one explores balancing trustworthiness and efficiency in AI Systems. We hope that this special issue will increase the community’s awareness of the importance of AI trustworthiness through data lineage.

查看原文本刊更多论文

值得信赖的人工智能和数据沿袭

人工智能的可信度是行业、政府和学术界最关注的问题。然而，人工智能及其模型的好坏取决于用于训练它的数据。数据沿袭可以通过多种方式进行跟踪，包括使用元数据，从数据的生成、使用、部署到验证。需要新的标准、蓝图、最佳实践和数据存储库来满足数据可信度的要求，例如可持续性、规模和响应能力，以及道德、多样性、公平性和包容性。在本期《IEEE互联网计算》特刊中，我们精选了三篇文章。第一个是关于可信的基于机器学习的应用程序的认证，第二个是关于深度学习中的数据和配置差异的主题，第三个是关于人工智能系统中可信度和效率的平衡。我们希望这期特刊能够通过数据谱系提高社区对人工智能可信度重要性的认识。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

IEEE Internet Computing 工程技术-计算机：软件工程

CiteScore

7.60

自引率

0.00%

发文量

审稿时长

6-12 weeks

期刊介绍： This magazine provides a journal-quality evaluation and review of Internet-based computer applications and enabling technologies. It also provides a source of information as well as a forum for both users and developers. The focus of the magazine is on Internet services using WWW, agents, and similar technologies. This does not include traditional software concerns such as object-oriented or structured programming, or Common Object Request Broker Architecture (CORBA) or Object Linking and Embedding (OLE) standards. The magazine may, however, treat the intersection of these software technologies with the Web or agents. For instance, the linking of ORBs and Web servers or the conversion of KQML messages to object requests are relevant technologies for this magazine. An article strictly about CORBA would not be. This magazine is not focused on intelligent systems. Techniques for encoding knowledge or breakthroughs in neural net technologies are outside its scope, as would be an article on the efficacy of a particular expert system. Internet Computing focuses on technologies and applications that allow practitioners to leverage off services to be found on the Internet.