Lill-DATA - A Framework for Traceable Active Learning Projects

Active Learning has become a popular method for iteratively improving data -intensive Artificial Intelligence models. However, it often presents a significant challenge when dealing with large volumes of volatile data in projects, as with an Active Learning loop. This paper introduces LIFEDATA, a Python-based framework designed to assist developers in implementing Active Learning projects focusing on traceability. It supports seamless tracking of all artifacts, from data selection and labeling to model interpretation, thus promoting transparency throughout the entire model learning process and enhancing error debugging efficiency while ensuring experiment reproducibility. To showcase its applicability, we present two life science use cases. Moreover, the paper proposes an algorithm that combines query strategies to demonstrate LIFEDATA's ability to reduce data labeling effort.