Designing & Delivering Machine Learning in Production
Sooner or later you probably will be working on a projects with words like “Data Science”, “Machine Learning”, “Big Data” attached to them. This talk will help you deliver these projects successfully by reviewing the key items you need to consider when incorporating analytical solutions into shipping products.
This talk will focus on answering these questions at scale:
- What are the most Agile ways to develop and deploy data- science heavy workloads?
- What is special about the computing environments doing data science or machine learning compared to non-science heavy workloads?
- What should we focus on when designing data transport and routing options in data science applications? How can we control costs and improve performance when doing this work?
- How do we setup test environments that verify the reliability of data science workloads?
The solutions we will be talking about will be a combination of tried-and-true techniques as well as very non-intuitive suggestions that make no sense in regular contexts. By the end of the talk you should be able to reason though the different design trade-offs when creating a sane and reliable production system.
About Husain Al-Mohssen
Husain's main focus is in the area of engineering science and it's application to create profitable products that serve 10's or 100's of thousands of users. Husain's work at MIT focused on extracting faint signals from super-computer scale gas simulations close to equilibrium. He later built on this background to start a user-facing email analytics company that serviced thousands of paying customers in near-real time. Husain has extensive software engineering experience both as a developer, maintainer and architect in the areas of enterprise software, high performance computing, as well as the “Big Data” domain. Husain is also an accomplished mechanical engineer, with many years of hands-on experience in the oil and power industries.
More About Husain »