Automating Machine Learning Systems: Lessons Learned

Ofer Ron (LivePerson)
Data Science
Location: 115
Average rating: ***..
(3.69, 13 ratings)

Many companies today deal with the need to deploy their hard earned machine learning knowledge for as many clients as possible. This challenge is impossible without robust automation of the machine learning systems.

However, machine learning automation is different than most engineering pursuits, since a critical part of the development takes place after deploying the system, where we learn if the ideas hatched in the lab actually scale and fit enough real world scenarios.

We’ll consider the different components needed to automate such a system, and various pitfalls encountered during this development, illustrated by war stories.

The talk will go over:

  • The philosophy of automatic configuration for scale
  • The components of a machine learning system: serving, learning, and monitoring
  • Data sanity and sanitation
  • The importance of internal A/B tests
  • The pros and cons of off the shelf technologies for machine learning
  • Sampling vs. learning over the entire corpus of data
  • The clash of cultures: engineers and data scientists developing together.
Photo of Ofer Ron

Ofer Ron

LivePerson

Senior data scientist and hacker at LivePerson, and tech lead for the research team. After getting a math Ph.D., I’ve spent the last seven years developing classification and recommendation systems over big data, taking them from the drawing board to production.

Comments on this page are now closed.

Comments

vladi Feigin
1-12-2014 10:40 CET

One of the best sessions in the conference.
Strongly recommend to take a look

Vasily Leksin
24-11-2014 10:34 CET

Hello! Thank you for your presentation, it was very helpful. Could you post the slides, please?