I-Bex Says BMVC 2026 Paper Accepted, Advancing Manufacturing AI Anomaly Detection
TECHWORLD ·
✦ AI Summary
I-Bex said its paper was accepted to BMVC 2026 after being recognized for research achievements in AI anomaly detection technology that enables continual learning of new data in a memory-constrained environment.
The accepted paper is titled "Memory-Bounded Continuation of Greedy Sampling for Continual Anomaly Detection."
The study focused on reducing memory usage in continual learning of vision-based anomaly detection models in manufacturing processes and edge AI environments.
I-Bex said on the second day after the announcement that it had been recognized for research achievements in AI anomaly detection technology that enables continual learning of new data in a memory-constrained environment. As a result, a paper in which I-Bex participated was accepted to "BMVC 2026."
The accepted paper is titled "Memory-Bounded Continuation of Greedy Sampling for Continual Anomaly Detection." BMVC is an international conference in the fields of computer vision and pattern recognition, hosted by the British Machine Vision Association (BMVA).
Following last year's paper acceptance at CVPR, I-Bex has continued its research in computer vision and AI with this year's paper acceptance at ICML and BMVC.
This study focused on reducing memory usage in the continual learning of vision-based anomaly detection models in manufacturing processes and edge AI environments. The environmental conditions involved the addition of new products and processes.
The existing approach is PatchCore-based anomaly detection. PatchCore is based on storing key features of normal data in a memory bank and is used for greedy sampling-based inspection.
However, in continual learning environments, the number of stored features increases as new data are continuously added. As a result, the problem of rising memory usage also emerges.
In response, I-Bex researchers proposed Memory-Bounded Continuation of Greedy Sampling. The technique aims to use both existing learning information and new data with limited memory resources, and it selectively retains the necessary information by reconstructing the distribution of previously stored features and newly introduced data within a fixed memory capacity.
The technology was designed to make it possible to add new process characteristics based on the retention of learning information from past data.
As a result, it can be used to continuously operate anomaly detection models on edge devices with limited memory resources, and it can also be used to continuously operate anomaly detection models on manufacturing lines producing multiple types of products.
I-Bex plans to apply the research findings to its own products, including its AI vision inspection platform, "AIVision," as well as expand its manufacturing AI business to include AI robotics, "AIVot," and the MLOps platform, "AIVOps." Seong Min-su, CEO of I-Bex, said the BMVC acceptance proved the academic rigor and field applicability of I-Bex's AI inspection technology, and added that the company plans to integrate the research results into its own platforms such as AIVision to help solve problems in manufacturing sites.
Source: TECHWORLD · Kim Seung-gi
Original: https://www.epnc.co.kr/news/articleView.html?idxno=406397
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.