This 3 hour instructional workshop will focus on the question of evaluating and interpreting deep neural networks (DNNs). We have previously organized workshops on deep learning at MedInfo 2015 and AMIA 2017. We have developed deep learning models for both biomedical and open-domain problems, and have introduced new methods for evaluating deep learning. In this workshop we will review deep learning and traditional methods for evaluation and interpretability, and introduce new methods for opening the black box of neural networks. Participants in the workshop will be able to experiment with evaluation and interpretation methods using models built for biomedical tasks in computer vision (e.g. image detection) and natural language processing (e.g. question answering).

Learning Objective 1: After participating in this session, the learner should be better able to understand how deep neural networks are trained, and understand steps taken to reduce the risk of overfitting models.

Learning Objective 2: After participating in this session, the learner should understand current methods for visualization and interpretation of deep learning models.


John Lalor (Presenter)
University of Massachusetts

Abhyuday Jagannatha (Presenter)
University of Massachusetts

Hong Yu (Presenter)
University of Massachusetts Lowell

Presentation Materials: