Abstract
ML-based applications already play an important role in factories in areas such as visual quality inspection, process optimization, and maintenance prediction and will become even more important in the future. For ML to be used in an industrial setting in a safe and effective way, the different steps needed to use ML must be put together in an ML pipeline. The development of ML pipelines is usually conducted by several and changing external stakeholders because they are very complex constructs, and confidence in their work is not always clear. Thus, end-to-end trust in the ML pipeline is not granted automatically. This is because the components and processes in ML pipelines are not transparent. This can also cause problems with certification in areas where safety is very important, such as the medical field, where procedures and their results must be recorded in detail. In addition, there are security challenges, such as attacks on the model and the ML pipeline, that are difficult to detect. This paper provides an overview of ML security challenges that can arise in production environments and presents a framework on how to address data security and transparency in ML pipelines. The framework is presented using visual quality inspection as an example. The presented framework provides: (a) a tamper-proof data history, which achieves accountability and supports quality audits; (b) an increase in trust by protocol for the used ML pipeline, by rating the experts and entities involved in the ML pipeline and certifying legitimacy for participation; and (c) certification of the pipeline infrastructure, the ML model, data collection, and labelling. After describing the details of the new approach, the mitigation of the previously described security attacks will be demonstrated, and a conclusion will be drawn.
Subject
Fluid Flow and Transfer Processes,Computer Science Applications,Process Chemistry and Technology,General Engineering,Instrumentation,General Materials Science
Reference52 articles.
1. Tsai, C.W. (2021). Toward blockchain for intelligent systems. IEEE Consum. Electron. Mag.
2. A comprehensive survey of clustering algorithms: State-of-the-art machine learning applications, taxonomy, challenges, and future research prospects;Ezugwu;Eng. Appl. Artif. Intell.,2022
3. Machine Learning Development Audit Framework: Assessment and Inspection of Risk and Quality of Data, Model and Development Process;Stodt;Int. J. Comput. Inf. Eng.,2021
4. Management Analysis of Industrial Production Losses by the Design of Experiments, Statistical Process Control, and Capability Indices;Bounazef;Open J. Bus. Manag.,2014
5. Lockey, S., Gillespie, N., Holm, D., and Someh, I.A. (2021, April 01). A Review of Trust in Artificial Intelligence: Challenges, Vulnerabilities and Future Directions. Available online: https://aisel.aisnet.org/hicss-54/os/trust/2/.
Cited by
5 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献