When you have the tools to quickly test any number of ideas for building a predictive model, new ideas from just about anyone can become valuable input.
After two weeks of competition in the Battle at the Nostradome, the leading teams are finding that Aspire's machine learning tools like A2i and PRDXT allow people from a much wider variety of disciplines to make valuable contributions for predictive model building.
"If you're able to synthesize the technical approach to the conceptual, anyone can engage," Aspire's Entrepreneur in Residence and Wylei CEO Mike Monteiro says. He's the leader of Team Pluto, the winners of the first and second weeks of the competition to build a model that predicts glucose levels for diabetes patients. The rest of Monteiro's team are from Aspire's finance department and the market research department.
Monteiro is an expert when it comes to machine learning and predictive software, but thanks to A2i he was also able to easily utilize input from others with no experience in predictive modeling.
It was a winning combination—beating out other teams full of data scientists, software engineers, and others with the technical know-how.
"We made a list of new things to try, and A2i simplified the process," Monteiro says.
If you only have one idea from one person, it's easy to test it out on your own, Monteiro says, but when you have ten people giving you ten ideas about predicting glucose levels, the problem becomes much more complex. Those ideas can interact with each other in an immense field of possible combinations, which is where A2i comes in.
"A2i lets you define a massive combinatorial search space and it says, ‘no sweat, I got this,'" Monteiro says. "It helps figure out what combinations of ideas are the best combinations."
So what exactly is A2i?
Essentially, Mike Monteiro says, to use the simplest terms: it's a search tool.
For those unschooled in artificial intelligence, search and optimization is a subfield of AI where machines really excel.
Search tools can be used to sift through an astronomical number of possibilities very quickly, allowing us to arrive at answers to problems that would otherwise be impossible with human reasoning alone.
But what's unique about Aspire's approach to search and optimization is that we're using it to find the best possible approach to machine learning (another subfield of AI); or in other words, we're using one category of AI to modify another category of AI. A process that could be called meta-artificial intelligence.
In the same way that humans use different parts of the brain in combination, we're using different subfields of AI in combination with each other to create systems that can easily adapt to new information.
Although the approach may not be unique from an academic perspective, Monteiro says, Aspire is one of the first to operationalize it.
The goal, Monteiro says, is to arrive at a process that will be easily transferable to many different problems, well beyond the specific problem of predicting glucose levels for diabetes patients.
"The vision is that you can leverage the power of machine learning and the scale of the cloud to build the next generation applications without having to be a data scientist," Moteiro says.
And judging by Pluto's success and their ability to engage people from other nontechnical disciplines, we're making progress to realizing that vision.