📑Case studies

Selected case studies

Topic	TLDR	Report
Anomaly Detection	AL reached same accuracy as random sampling with up to 69% less data	https://docs.google.com/document/d/1YTNAwf-wveTsZH9KbXB06oLky4Djic87BdhpaDrgK9o/edit#heading=h.w3cedc6xt339
Semantic Segmentation	AL can do better than random with 70% less data	https://docs.google.com/document/d/1aB_lzishbt3ChdQIcfCWntXz6ZG1AXAly3sAgS06Vsc/edit?usp=sharing
Classification	AL can drive global improvement as well as targeted local improvements	https://docs.google.com/presentation/d/1X8kQwtUSm9XvjYwnuvRfDQ10rE53IYxvwe60KuVxCQM/edit#slide=id.p
Object Detection	AL can be used to sample up to 16x more errors than random sampling
Classification	AL can correct data drift and improve a accuracy up to 22% in a single iteration