Put open source to work
July 16–17, 2018: Training & Tutorials
July 18–19, 2018: Conference
Portland, OR

Going deep: A study in migrating existing analytics to deep learning

Ryan Roser (Refinitiv)
11:50am12:30pm Thursday, July 19, 2018
Artificial intelligence
Location: D137/138
Level: Intermediate
Average rating: *****
(5.00, 1 rating)

Who is this presentation for?

  • Data scientists and product managers

Prerequisite knowledge

  • Familiarity with creating products powered by analytics
  • A basic understanding of deep learning

What you'll learn

  • Understand how deep learning can improve existing analytics products
  • Learn the challenges involved in updating a model to use deep learning

Description

The San Francisco Innovation Lab at Thomson Reuters creates quantitative models for investors. Following the financial crisis, it developed the first commercially available credit risk model to measure corporate financial health by systematically evaluating the language used in news, conference call transcripts, financial filings, and analyst research. Six years later, the company is updating it to take advantage of deep learning.

While deep learning promises breakthroughs in the ability to model complex data and power new products, when does it make sense to migrate existing analytics to use deep learning, and what advantages does it offer in practice? Furthermore, what are the trade-offs?

Ryan Roser answers these questions and describes the process of updating an existing quantitative model to use deep learning.

Topics include:

  • How preexisting model requirements can inform deep learning architecture
  • The role of feature engineering in deep learning
  • Opportunities to expand existing text-based analytics to support non-English content
  • Cost and complexity around developing a deep learning model versus updating traditional analytics
  • Ways to deal with interpretability when using black box models
Photo of Ryan Roser

Ryan Roser

Refinitiv

Ryan Roser is the director of data science and text analytics within the San Francisco Innovation Lab at Refinitiv, where he develops quantitative models and predictive analytics for investors and works with unstructured text to identify new trends and insights. Previously, Ryan was a principal quantitative research analyst at StarMine, where he developed a first-of-its-kind text-based corporate credit risk model. Ryan lives in Portland, Oregon. He enjoys gardening and raising chickens.