Explainable Deep Machine Learning Methods in Detecting Diseases

Weiqing Gu, DASION Corp. ABSTRACT: This talk will explore the use of Geometric Unified Learning (GUL) technology to create explainable deep learning methods. GUL, consisting of reusable building blocks, addresses several challenges in deep learning in detecting diseases by providing increased value to end-users and clinicians. It saves time and resources in data processing, fitting data before it is input to a deep neural network to avoid overfitting, providing transparency and trustworthy solutions which are crucial in medicine that can easily be understood by data-to-decision makers and doctors, making data analysis interesting for software developers, scientists, clinicians, and engineers outside of the machine learning field, producing flexible, robust and agile solutions for debugging, and saving money and time. Saving doctor’s time would in turn save lives. Rooted in differential geometry, GUL creates appropriate local coordinate systems, Riemannian metrics, transformations, and geodesics to identify data invariants and intrinsic patterns, resulting in highly interpretable results. The GUL tools have capabilities of vectorizing data, compressing data, searching and learning simultaneously. When the technology makes predictions, it will show the user exactly which data points are responsible for those predictions. This research addresses several technical challenges in deep learning such as garbage-in producing garbage-out, needing large volumes of training data, long running times, unexplainable outputs, and excessive parameter tuning.