Patents are the currency of innovation, in software or any other industry. At FICO World 2016 in April, I explored how four patents we have recently been granted will enrich FICO’s product portfolio and, in turn, your business.
From cybersecurity, to payment card fraud, to marketing, companies want answers to the question, “How do we know when someone's behavior has changed in a significant way?” And subsequently, “How do we rank those changes to focus on the most important cases?”
FICO's recent patent on Collaborative Profiling and change-point detection provides an efficient new way to answer these questions. By distilling behaviors from a large group down to a few basic “archetypes” of behavior, we can effectively determine which persons have the biggest changes in archetype mixtures. That’s a powerful capability that can also help marketers to better understand changing customer preferences.
Structuring IoT Messages
For the past few years there’s been a lot buzz about the Internet of Things (IoT), composed of billions of devices that emit vast amounts of log lines in the shape of free-form text messages. Organizations would like to leverage this massive, yet poorly structured, information source in more automated and effective ways. At the outset, this requires structuring log messages into more regular categories before advanced downstream models for anomaly detection and prediction can apply.
One of our patent applications addresses this important problem by “Identifying Latent States of Machines Based on Machine Logs.” Here, we invented a novel way to group, cluster and ultimately structure machine log messages into sequences of states for processing by downstream analytic models. Key to this method: a fast streaming implementation that can run efficiently on the huge scale of IoT. We are excited for the potential of this development to improve the management and security of IoT devices.
Neural Networks in Data Governance
Model governance is a key challenge for any organization that uses predictive analytics for mission-critical tasks (like financial decisions). But how do we apply data science to the governance process itself? Whether it's data validation prior to certifying a model for production, or deciding which model is best, FICO's new Autoencoder Self-Diagnostic technology provides an answer.
Autoencoders are a type of neural network that can be trained without hand-labeled training sets. Here, the autoencoder learns to mimic its input data through a compressed representation that captures complex interactions between different factors. This can be used to understand data that moves away from what the model was trained on, or to identify segments of data that are not well represented in the model.
New Optimization Technology
How do you effectively optimizing offers made to customers and prospects? This question affects everything from marketing offers to financial decisions such as setting a credit card customer’s credit limit.
FICO recently patented a methodology for automating the complex analytic tasks required to optimize an offer strategy. We approach this in a novel way, creating a dynamic system capable of self-learning the optimal strategy quickly, after a few revolutions of a test-and-learn cycle. This allows companies to respond much more quickly to market conditions and customer demands.
These four patents can each be applied to many problem areas, and our analytics team is looking forward to rolling them into our newest products. Follow me on Twitter @ScottZoldi.