Loading…
Thursday, October 3 • 2:50pm - 3:00pm
Lightning Session- Deep Learning Model Development in the AWS Ecosystem

Sign up or log in to save this to your schedule and see who's attending!

Deep Learning Model Development in the AWS Ecosystem
Deep learning has the potential to help automate many data analysis problems within drug discovery and even more broadly scientific research. In areas such as image analysis where the previous solutions worked only moderately well at best, deep learning can now be applied to solve problems completely and with high performance. Unfortunately, the process of developing deep learning models is a highly custom one, with each use case requiring its own set of data management software and models trained on specific data. There is however quite a bit of commonality that can be abstracted out of the process of deep learning model development and reused across different problems. Amazon Web Services (AWS) is a cloud-based ecosystem of on demand computing resources that provides tools in support of this effort.

As the deep learning model development process requires the manipulation and management of large amounts of data and specialized computing resources, AWS offers an ideal environment for the entire process. We make heavy use of not only the core services inside AWS such as S3 and EC2, but also the fully managed service Amazon SageMaker, which is a complete and customizable machine learning model development tool. SageMaker allows us to collect and label data, and then train and deploy models all in one end to end workflow. This represents a large time savings versus the alternative of setting up our own custom software and infrastructure for each project. SageMaker also provides us with a system to track datasets and models, which saves time for tasks like transfer learning and helps bring order to the whole model development process.

 Adopting and leveraging tools like SageMaker and AWS to develop custom models in support of scientific research, changes what was a complex and time intensive process, to one that supports more rapid and scalable implementation. This changes the value proposition in adopting deep learning for scientific research.

Speakers
avatar for Steven Van Buskirk

Steven Van Buskirk

Engineer, Amgen
Steven Van Buskirk is an engineer with over 8 years of biotech experience. At Amgen he is focused on developing deep learning models in support of Discovery Research. Prior to Amgen Steven worked at several biotech start-ups and research labs, including QuantaLife (now Bio-Rad), InDevR... Read More →

Chairs
avatar for Mike Berke, MS

Mike Berke, MS

Director of Research and Automation Technologies, Amgen
Mike is the Director of the Research & Automation Technologies (R&AT) group at Amgen, where he has been for the last 16 years. The R&AT group focuses on developing novel devices and software tools to improve research processes.


Thursday October 3, 2019 2:50pm - 3:00pm
Theater Ballroom (5th Fl) - Courtyard Boston Downtown 275 Tremont Street, Boston, MA 02116