Machine Learning in the Industry

In this section you can find all the information for the session Machine Learning in the industry scheduled on 6/9/2018.


Title: Machine Learning and Big Data in the hands of customers

Speaker: José Antonio Rodríguez Serrano, BBVA Data & Analytics

Time slot: 14:30-15:15

Abstract:

If we ask you which industries have lots of data and leverage machine learning for competitive advantage, you will probably think of search engines, ecommerces, telcos or social networks. And yet, you might not be aware of it, but a new older player is coming (spoiler: banks). Back some years now, BBVA realized that they had the data to be a major big data player: not only for knowing people purchasing behavior, but also their financial health, the relationships among companies (client-supplier, financial risk, etc), web interactions, apps, etc. So the question is, can a bank become a data and algorithmic driven company?

BBVA Data & Analytics, a center of excellence in financial data analysis within the BBVA group, has delivered in the last years several data engines that provide the intelligence behind digital banking products, in complex setups that involve cutting-edge technologies working in coordination with legacy systems. In this talk we will revisit our experience conceiving, designing, developing and delivering these highly complex products. We will discuss how we organize ourselves to spearhead the transformation of BBVA, focusing on three main stages in the lifecycle of digital products: inception of an idea using design fiction; development with an agile-based software process tailored for data science, in close collaboration with other business areas; and delivery in coordination with existing systems to put our data-enabled digital products in the hands of customers.


Title: Data-Efficient Reinforcement Learning

Speaker: Haitham Bou Ammar, Prowler.io

Time slot: 15:15-16:00

Abstract:

Though successful in numerous applications, current reinforcement learning techniques suffer from high computational and sample complexities. This limits their application in real-world scenarios, where environmental interactions are expensive. At PROWLER.io, we are developing next-generation reinforcement learning algorithms that are efficient, scalable, and robust. To do so, we draw upon a variety of methodologies from different fields, including probabilistic modelling, game theory, and optimisation. In this talk, I demonstrate how reinforcement learning can be made much more data-efficient. As an example, I present a result for controlling Montezuma's revenge in the order of thousands of interactions with the environment. Furthermore, I describe a model-based algorithm that controls continuous systems in tens of interactions.


Coffee Break

Title: Lynx: real-time accurate fraud detection over massive data

Speaker: Álvaro Barbero Jiménez, Instituto de Ingeniería del Conocimiento

Time slot: 16:30-17:00

Abstract:

In this talk we will present Lynx as a notable example of machine learning in industry. Lynx is a solution for real-time fraud detection in payments, a challenging problem due to a number of reasons:

  • Large volume of data: an average of 6 million operations per day.
  • Large class imbalance: 1 fraud for every 10.000 legit payments.
  • High frequency real time constraints: max. 50 milliseconds per prediction.
  • Constantly shifting fraud distributions.
Lynx solves these issues by implementing a classification model that is updated every day with new training data. The solution is made of a blend of "classical" technologies (C, shell script) with new ones (Python, Elastic) to guarantee real time performance. It is currently being used in 7 countries, analyzing 27 billion transactions per year, including every single transaction performed in Spain.

References: http://www.iic.uam.es/en/big-data-services/banking-environment/lynx/