Active learning

Last updated Apr 10, 2022 Edit Source

Active learning refers to algorithms that take an active role in the selection of which ex-amples are labeled. Active learning assumes that there is an ‘oracle’, such as a human expert, that can be queried to get ground-truth labels for selected unlabeled instances

• https://en.wikipedia.org/wiki/Active_learning_(machine_learning).
• There are situations in which unlabeled data is abundant but manually labeling is expensive. In such a scenario, learning algorithms can actively query the user/teacher for labels. This type of iterative supervised learning is called active learning. Since the learner chooses the examples, the number of examples to learn a concept can often be much lower than the number required in normal supervised learning. With this approach, there is a risk that the algorithm be overwhelmed by uninformative examples.

• Overview of Active Learning for Deep Learning
• #PAPER An open source machine learning framework for efficient and transparent systematic reviews (van de Schoot 2021)
  • #CODE https://github.com/asreview/asreview

• #PAPER Active learning literature survey (Settles 2010)
• #PAPER Active Learning for Convolutional Neural Networks: A Core-Set Approach (2018)
• #PAPER Rethinking deep active learning: Using unlabeled data at model training (Simeoni 2019)
• #PAPER Deep Batch Active Learning by Diverse, Uncertain Gradient Lower Bounds (2020)

• #PAPER Incorporating Expert Feedback into Active Anomaly Discovery (Das 2016)
  • http://web.engr.oregonstate.edu/~tgd/publications/das-wong-dietterich-fern-emmott-incorporating-expert-feedback-into-active-anomaly-discovery-icdm2016.pdf
• #PAPER Deep Active Learning for Anomaly Detection (Pimentel 2018)