ML is not enough: Decision automation in the real world
Who is this presentation for?
- Data scientists, machine learning engineers, CTOs, product managers, and those interested in building practical real-world AI software systems
With the recent proliferation of data in all aspects of the enterprise, the use of machine learning and predictive models has become pervasive across modern organizations. However, the real value of an AI system goes beyond data-driven predictions into full decision automation. Automating decisions in the wild requires careful consideration of implicit constraints and fuzzy business objectives to ensure that decisions are robust and consistent with the business goals.
Brian Keng takes a deep dive into leveraging modern data-enabled machine learning methods and traditional operations research optimization techniques for decision automation in the context of retail enterprise applications. He reviews the challenges of automating decisions ranging from specification of soft and hard constraints combined with fuzzy objectives. Borrowing techniques from other disciplines, such as operations research and mathematical optimization, Brian explores how to combine modern ML techniques with traditional operations research methods to solve two important problems in enterprise retail. The first problem details how you generate product recommendations that are consistent with business requirements, such as contractual obligations, diversity and serendipity, and incremental sales. The second problem describes how you solve the inventory management problem that takes into account various supply chain constraints while simultaneously leveraging data-driven ML techniques to optimize for overall financial impact. He explains some general strategies to ensure that data-driven predictions used in decision making are robust and consistent with business objectives and goals.
- A basic understanding of machine learning and mathematical optimization
What you'll learn
- Understand that automating decisions requires you to take into account constraints and objectives that are not readily available using machine learning methods
- Discover how modern machine learning methods can be combined with traditional operations research optimization techniques to build a robust decision automation system
Brian Keng is the chief data scientist at Rubikloud, where he leads a team building out intelligent enterprise solutions for some of the world’s largest retail organizations. Brian is a big fan of Bayesian statistics, but his main professional focus is building out scalable machine learning systems that seamlessly integrate into traditional software solutions. Previously, Brian was at Sysomos, leading a team of data scientists performing large-scale social media analytics, working with datasets such as the Twitter firehouse. He earned his PhD in computer engineering from the University of Toronto, during which time he was an early employee of a startup that commercialized some of his research.
For conference registration information and customer service
For more information on community discounts and trade opportunities with O’Reilly conferences
For information on exhibiting or sponsoring a conference
For media/analyst press inquires