Presented By
O’Reilly + Intel AI
Put AI to Work
April 15-18, 2019
New York, NY

Risk-free Deep Learning without Sacrificing Performance

Evan Sparks (Determined AI)
1:00pm1:40pm Thursday, April 18, 2019
Implementing AI
Location: Mercury Rotunda
Secondary topics:  Deep Learning and Machine Learning tools, Platforms and infrastructure

Who is this presentation for?

Data Scientists, Machine Learning Engineers, Director of Machine Learning



Prerequisite knowledge

Audience members should be familiar with the basics of deep learning: how does training happen, how are models initialized, how do they get updated over time, and how much time and effort goes into training a deep learning model. Hands-on experience with common application frameworks (TensorFlow, Keras, PyTorch) and source controls systems (git) a plus but not mandatory.

What you'll learn

Audience members will learn what what inputs are necessary if they want reproducibility in their model development workflow. They'll learn how to adapt their processes to track these inputs and incorporate them in their development process. They'll learn how, without care, insisting on reproducibility can slow down computational performance. Finally, they'll learn how to maintain high performance while retaining reproducibility.


Building applications powered by Deep Learning is hard. It’s particularly hard in an enterprise setting when reproducibility is of paramount importance: it’s difficult to justify shipping a model to production when you can’t be sure you’ll be able to build it again. Achieving reproducibility during deep learning model development comes at a cost: it can be difficult to maintain a high degree of computational performance during the model development lifecycle while retaining the benefits of reproducibility. In this talk we’ll describe the key ingredients of reproducible deep learning models in an enterprise setting. We’ll then focus on how to maintain a high degree of resource utilization and throughput through workload-aware cluster resource orchestration techniques.

Photo of Evan Sparks

Evan Sparks

Determined AI

Evan Sparks is cofounder and CEO of Determined AI, a software company that makes machine learning engineers and data scientists fantastically more productive. Previously, Evan worked in quantitative finance and web intelligence. He holds a PhD in computer science from UC Berkeley, where, as a member of the AMPLab, he contributed to the design and implementation of much of the large-scale machine learning ecosystem around Apache Spark, including MLlib and KeystoneML. He also holds an AB in computer science from Dartmouth College.

Leave a Comment or Question

Help us make this conference the best it can be for you. Have questions you'd like this speaker to address? Suggestions for issues that deserve extra attention? Feedback that you'd like to share with the speaker and other attendees?

Join the conversation here (requires login)