Dataperformers is selling science as a service

If you go back 15 years or 20 years, we had some of the ideas that are proving to be really good now, but we just didn’t have the computational capacity, nor the amounts of data that we have now at our disposal.

Businesses come to Dataperformers with problems they think artificial intelligence can solve. The company then researches the problem and develops an AI-based solution.

Mehdi Merai, Dataperformers’ co-founder and CEO, calls it “science-as-a-service.”

“We are a company that solves problems through applied research work in artificial intelligence,” Merai said.

Clients include the Desjardins Group.

Dataperformers is “working to dive into our ‘data lake’ and discover what we call the ‘unknowns,’ ” Deborah Hayek, the head of research and development at Desjardins Lab, said in a June interview. “One big advantage of AI is it’s able to parse through data in an incredibly fast way.”

AI systems can go though vast amounts of data — like Desjardins’ “data lake” (its repository of raw data) — and AI is also able to identify patterns and trends that humans wouldn’t be able to find, she said.

In addition to its service business, Dataperformers is developing two products that it plans to sell. One is a search engine for video. Right now, search engines look only at the metadata of online videos — things like the title, runtime and who uploaded it to a specific website. They can’t know what’s shown in the video itself.

Dataperformers’ video search engine uses a combination of computer vision and artificial intelligence to find specific content in video.

The idea, Merai said, is that companies would use the search engine to identify videos where their products appear. They could then market those products directly to people watching the video.

It’s not easy to train a computer to recognize a specific product in a video.

Merai uses the example of a Tshirt. Using traditional AI training methods, Dataperformers would have to “show” the system thousands of images of that T-shirt so it could distinguish that specific shirt from all others.

But there’s no way for the company to train its system on every single product a user might search for.

That’s the challenge — teaching the system how to match a pattern without training the AI the classic way, Merai said.

To solve that problem, the company is taking a different approach, called reinforcement learning, where a machine learns on its own based on feedback.

Merai gives the example of a robot that’s placed in a room and needs to get to a specific place.

“The robot will explore different actions and get the feedback and after that it will look for the right sequence of actions that bring it to the right place,” he said.

The company’s other product is intended to help financial services professionals.

Traditionally, these professionals develop hypotheses and then look for data that support, or disprove, those hypotheses.

Dataperformers wants to flip that.

“The machine will search for correlations, hidden phenomena, weak signals and try to build hypotheses to push to the user,” Merai said.