A new arXiv paper introduces TypiCore, a method that combines active learning queries with class-incremental learning for time series data. The approach targets settings such as healthcare and manufacturing, where models must adapt to distribution shifts as new classes arrive over time. The work is posted as a cross-listing replacement on arXiv and has not yet undergone peer review.
A new preprint introduces BioDCASE, an active learning approach aimed at reducing the labeling burden in passive acoustic monitoring for ecology. The work targets the high cost of annotating large audio datasets by selecting which recordings are most worth labeling. It is published on arXiv under the cs.LG category.
A new arXiv preprint argues that most machine learning approaches to clinical diagnosis assume fully observed, static datasets, which does not match how clinicians reason sequentially while weighing resource constraints. The authors propose a method that actively chooses which diagnostic test to order next, aiming to reach a diagnosis sooner with fewer tests. The work is presented as a replacement version of the paper (v5).
A new arXiv paper examines whether internal transformer activations can guide the selection of in-context examples for large language models. The authors note that earlier work applied active learning to in-context sample selection, but without drawing on recent findings about how activations encode information. The study reports mixed results and lessons from testing this activation-based approach.
A new study presents an active learning framework designed to connect laboratory-scale results with real decisions about which critical-materials recovery processes to scale up. The method accounts for product requirements, process costs, and scale effects, drawing on records from Pacific Northwest National Laboratory's critical-element recovery database.
A newly posted arXiv paper explores combining active learning with the lottery ticket hypothesis, asking whether sparse subnetworks that match dense-model accuracy can be found while also reducing labeled data needs. The authors frame this as achieving both sparsity and sample efficiency within a single training process rather than running separate search stages. The work appears in both the machine learning and AI categories of arXiv.