Glossary term
Glossary term
Training and Fine-Tuning
A training approach in which the algorithm chooses some of the data it learns from. Active learning is particularly valuable when labeled examples are scarce or expensive to obtain. Instead of blindly seeking a diverse range of labeled examples, an active learning algorithm selectively seeks the particular range of examples it needs for learning.
Created for this library
A law firm's document classification team uses active learning so junior associates only label the documents the model is least certain about, cutting labeling spend roughly in half.
A medical device company runs an active learning loop where a radiologist annotates only the lung scans the segmentation model flags as ambiguous.
An autonomous driving team uses active learning to prioritize labeling the dash-cam frames where its detector confidence is lowest, focusing reviewers on hard cases.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License