Millimeter-Scale Computers: Now With Deep-Learning Neural Networks on Board (16 February 2017 12:34)
Date: 1 December 2016 10:34
That time of year is upon us again – the one where we suddenly realize we're running out of time to buy gifts for everybody. But what to buy? That eternally tricky question may soon may soon be outsourced to artificial intelligence: by answering a few questions about the recipient, a service called ebo box is designed to pick out the right gift for them using deep learning algorithms to scan market data and look for connections that we mere humans might miss.
Neural networks – AI systems that "learn" over time – have proven themselves useful for processing huge amounts of data, with applications in areas as broad as farming, genetics and art. ebo Innovations trained its network on a database of 21 million people, who anonymously answered a survey about their interests, brand loyalties, what kinds of products they do or don't like, and more general preferences like their favorite color.
The system uses that information to help gift-givers find the right present. Users punch in a few details about themselves and the person they're buying for, from the obvious questions like age, gender, location and interests, to more specific and personal points, like how extroverted or creative they may be. A price limit can be set too, to keep suggestions reasonable.
ebo's system then builds a profile of the recipient and scans its database to find connections that may not be obvious to the giver: does Great Uncle Fred, for example, drink Coke, love hiking and favor the color green? Based on the data, he may be far more inclined to buy a certain type of smartphone.
"Artificial intelligence can uncover linkages that humans can't," says Marwan Sledge, ebo founder and CEO. To get the AI-enhanced gifts into the hands of the giver and then the recipient, ebo box seems to have taken up a Loot Crate-style model. When ordering, people can set the service to be a once-off transaction or a regular subscription, and the user doesn't actually know what the gift is until they open the box when it arrives at their door.
Currently, the process takes about a day for the system to process the data, but as it develops, ebo hopes to tighten that up to a few hours. Eventually, the company also wants to wean the network off the third-party survey data and onto its own database of information collected from ebo box participants, which includes how well the gift was received. That feedback should be able to help the system learn from its successes or mistakes, and as neural networks are wont to do, get better at its job over time. For now, ebo box is still under development, aided by Nvidia's Inception program.