Overview

It is estimated that 100 million to 2 billion human genomes will be sequenced by 2025. In addition to genomic data, each person will produce terabytes of microbiomic, physiologic, anatomic, biologic, demographic and environmental data. The modalities and complexities of this data are manifold. Researchers need to understand it both at an individual and population level which presents one of the most complex technological challenges of our generation. Improving our understanding of this data has potential groundbreaking impacts in the prevention, diagnosis and treatment of diseases as well as developing better drugs and food sources.

Advancements in Artificial Intelligence (AI) in recent years make it now possible to solve very complicated tasks like never before. Specifically, deep learning has shown great promise at solving long-standing challenges in AI such as visual understanding, speech recognition and natural language processing. The progress in deep learning is possible due to the (1) advancements in computational power, mainly parallel processing, (2) invention of efficient algorithms and tools and (3) availability of mass data that makes using complex neural network models possible.
In this talk, I will discuss deep learning and illustrate technological breakthroughs that enable its usage on complex and large scale problems. As well, I will showcase how deep learning is solving problems in the life sciences, including applications in digital pathology, ultrasound imaging and DNA sequencing. I will also talk about how a large-scale infrastructure and following best practices can help machine learning teams achieve better results when applying deep learning methods.