The brother-in-law with the personality of a doorknob. The teenager who lives to text. The person who already seems to have it all.
There’s always at least one on everybody’s holiday shopping list: the person who’s impossible to buy for. Now, GPU-accelerated deep learning promises to wrap up gift-giving faster than the time it takes to drive to the mall.
ebo box, a San Jose-based startup, says it can help find gifts for even the most perplexing recipients by learning about the gift-givers and recipients, and combining that with data it’s collected about general user preferences in the market. The company calls it “smart gifting.”
How Deep Learning Gift-Giving Works
After setting up a profile, ebo gift-givers tell it how much they want to spend and answer a series of questions about themselves and the person receiving the present. Besides establishing basics like the recipient’s gender, age and location, ebo asks about characteristics such as the recipient’s interest in outdoor activities, whether the giftee is introverted or extroverted, and whether the person is strictly logical or leans toward the creative side.
Next, ebo uses deep learning to find connections between user preferences and products that wouldn’t be obvious to a human shopper. Because ebo doesn’t yet have user data of its own, it trained its neural networks on an anonymized database of 21 million survey participants that includes information like products and brands they like and dislike, personal interests and even favorite colors.
“Artificial intelligence can uncover linkages that humans can’t,” said Marwan Sledge, ebo founder and CEO. Sledge said his algorithms might find hidden connections showing, for example, that people who drink one type of soda, prefer the color green and love hiking, are also 52 percent more likely to buy a specific brand of smartphone.
Finding those kind of connections in massive amounts of data requires heavy-duty computing, Sledge said. In addition, as more people answer survey questions, the company periodically retrains its neural network to reflect the added information. When ebo has its own users, it will phase out the survey data and retrain its neural networks based on the information they enter and on how well presents were received.
“That’s why we started using NVIDIA GPUs and that’s why we use deep learning,” he said.
Using NVIDIA GPUs and CUDA, the company’s reduced computation time from four or five days to a single day. And it’s working to shorten that to a few hours.
ebo recently joined our Inception program, which supports innovation and growth of startups that are driving breakthroughs in AI and data science.
Why Deep Learning Gift-Giving Was Born
Sledge said he was inspired to create ebo because of his own troubles with gift giving.
“I am terrible at choosing gifts, and I get bored of giving gift cards,” he said. After conducting market research, Sledge found he wasn’t alone.
He began to develop the deep learning smart gifting algorithms that became ebo in his spare time while working at Apple.
Users choose the price they want to pay per gift. The minimum gift is $20. The maximum? The sky’s the limit.