Uncertainty-Driven Active Learning in Deep Hybrid Models
Supervision: Prof. Dr. Nassir Navab,
Dr. Shadi Albarqouni
Abstract
The availability of large amounts of quality labeled data is a fundamental challenge of modern supervised learning. semi-supervised learning techniques try to leverage the use of a small amount of labeled data to train a large dataset of unlabeled data. However, choosing which labeled data to use is not usually addressed. In this work, we explore the use of only uncertain data samples in training. We cluster our unlabelled data using deep Gaussian mixture model. Uncertainty is then modeled using Monte-Carlo dropout, and uncertain data points are used to train our classifier.
We validate our work on two toy datasets as well as a real-world medical application.
Requirements:
- Good understanding of statistics and machine learning methods.
- Very good programming skills in Python & TensorFlow? / PyTorch?
Location:
Literature
Resultant Paper