Implementation and Validation of non-semantic Out-of-Distribution Detection on Image Data in Manufacturing

Date

2023-01-03

Contributor

Advisor

Department

Instructor

Depositor

Speaker

Researcher

Consultant

Interviewer

Narrator

Transcriber

Annotator

Journal Title

Journal ISSN

Volume Title

Publisher

Volume

Number/Issue

Starting Page

6675

Ending Page

Alternative Title

Abstract

Small changes in the production environment can have a negative impact on the performance of machine learning models. This study investigates the feasibility of various methods for detecting non-semantic Out-of-Distribution (OOD) cases in input images, which could be caused by hardware-side malfunctions, such as a defective camera flash. For this purpose, we design four experiments based on a real-world computer vision use case to simulate hardware problems that may occur in manufacturing and verify the performance of the various methods for detecting OOD cases. Furthermore, we explore the optimal sample size of input data to ensure that OOD cases can be found efficiently and successfully. The experimental results show that the tested methods can effectively and correctly detect the presence of non-semantic OOD data. The next step is to focus on securing ML models to identify malignant OOD cases, which negatively affects the performance of deep learning models.

Description

Keywords

Cybersecurity and Software Assurance, ai security, computer vision, manufacturing, out-of-distribution detection

Citation

Extent

10

Format

Geographic Location

Time Period

Related To

Proceedings of the 56th Hawaii International Conference on System Sciences

Related To (URI)

Table of Contents

Rights

Attribution-NonCommercial-NoDerivatives 4.0 International

Rights Holder

Local Contexts

Email libraryada-l@lists.hawaii.edu if you need this content in ADA-compliant format.