Feature Labs Inc., a startup founded in 2015 at the Massachusetts Institute of Technology’s computer science department, today hit the scene with $1.5 million from investors to change how organizations build artificial intelligence models.

The company is focusing its attention on the labor-intensive element of AI projects known as feature development. The process involves identifying the variables, or “features,” that a machine learning model should consider when making decisions. In a computer vision context, for example, the features might be common visual characteristics of the objects an AI is designed to identify.

The task can requires a massive time investment thanks to the large number of attributes that need to be accounted for in an enterprise-scale machine learning project. To streamline the task, Feature Labs offers an automation platform based on an algorithm called Deep Feature Synthesis that Chief Executive Officer Max Kanter developed while at MIT.

The software can take over much of the hands-on work required to identify relevant data attributes for an AI. According to Feature Labs, its platform can find patterns 10 times more quickly than traditional methods.

The startup is initially applying this automated approach to finding features within relational and transactional data. The two categories cover a wide range of record types ranging from user behavior information, such as website activity logs, to bank payments.

Feature Labs said several large organizations have already adopted its platform to scan their data for patterns. The startup’s user base includes NASA, the Defense Advanced Research Projects Agency (better known as DARPA), Kohl’s Corp. and Spanish banking giant BBVA SA.

BBVA is applying Feature Labs’ software for fraud detection. According to the startup, BBVA has used the technology to develop models for catching illicit transactions that proved to be 54 percent more accurate than its existing system.

Feature Labs is one of several companies working to automate AI development. Among the other players in this segment is Google LLC, which recently launched a cloud service called AutoML that enables users to train machine learning models for specific tasks with drag-and-drop controls. The offering initially targets computer vision applications, with support for more use cases set to arrive further down the road.

Image: Unsplash