Active Learning system as an AI tool in manufacturing environments explained

The Jožef Stefan Institute (JSI) develops Active Learning systems as part of an EU-funded project STAR – Safe and Trusted Human Centric Artificial Intelligence in Future Manufacturing Lines. The project addresses the challenge of trustworthiness and applicability when replacing human tasks in dynamic operations by developing new technologies that enable the implementation of standards-based, safe, reliable, and trustworthy human-centric AI systems in manufacturing environments. The STAR project aims to explore and integrate leading AI technologies such as active learning systems, which will be shortly explained in this article.

Active Learning: A Powerful Tool For Accelerated Acquisition Of Knowledge

Methods and approaches for active learning (AL) are typically the natural approach to provide human-in-the-loop functionalities (models that require human interaction) in advanced AI systems. Active Learning usually attempts to improve learners’ performance by asking questions to an expert to obtain labels for data instances. Since users often do not wish to provide information and feedback, AL is used to identify a set of data instances on which user input provides the system with the most valuable information.

Collecting feedback from prediction explanations can be implemented using a framework of three components: a prediction engine, an explanation engine, and a feedback loop to learn from users. JSI extends this approach to collect feedback on predictions, explanations of predictions, and decision options that are recommended to users.
Moreover, JSI has developed a system that is based on semantic technologies considering ontology concepts that are generic and can be ported to multiple use cases and can capture and encapsulate complex knowledge. To develop decision workflows that are displayed through an interactive user interface demand prediction models, explainable AI, a recommendation system for decision making and a knowledge graph are integrated. User feedback is collected on the predictions, the explanations of the predictions, and the decision options displayed to the users.