DEAL Active Learning

The paper “DEAL: Deep Evidential Active Learning for Image Classification” was accepted for publication at the 19th IEEE International Conference on Machine Learning and Applications (ICMLA)

  • Date: 27.01.2021
  • The paper “DEAL: Deep Evidential Active Learning for Image Classification” of our former master student Patrick Hemmer (DSI of IISM/ KSRI) together with Niklas Kühl and Jakob Schöffer was accepted for publication at the 19th IEEE International Conference on Machine Learning and Applications (ICMLA).
    In the paper, the authors propose a novel AL algorithm (DEAL) that efficiently learns from unlabeled data by capturing high prediction uncertainty. By replacing the softmax standard output of a CNN with the parameters of a Dirichlet density, the model learns to identify data instances that contribute efficiently to improving model performance during training.In several experiments with publicly available data, it is demonstrated that DEAL consistently outperforms other state-of-the-art Active Learning approaches. The method can be easily implemented and does not require extensive computational resources for training.
    By applying DEAL to a real-world use case in the field of automated pediatric pneumonia detection in chest X-ray images, the performance of the method can be confirmed. .
    Feel free to contact the authors for more information.
    Link to paper: https://arxiv.org/abs/2007.11344
    Link to conference: https://www.icmla-conference.org/icmla20/