R2 uses big data to develop AI solutions for enterprise customers | Tech Industry

Artificial intelligence is the hot new thing in enterprise. A whopping 80 percent of businesses have some form of machine learning in production today, according to a study commissioned by Teradata, and fully 21 percent have ongoing AI pilot projects.

But what about the businesses starting at the ground floor? Those without AI expertise might turn to R2.ai, a three-year-old Silicon Valley-based startup providing tools that help customers develop AI solutions.

“We’re the AI that helps create AI,” CEO Yiwen Huang told VentureBeat in an interview. “Our goal is to make data accessible for all [and to] drive mass adoption across different industries.”

Its flagship product is R2 Learn, which Huang compares to Google’s AutoML: a suite of machine learning services that enable developers to train high-quality models. Unlike AutoML, though, which takes a deep learning approach to AI, R2’s bread and butter is predictive analytics.

R2 Learn is to an extent automated, but R2 makes available a small army of consultants to guide businesses through the model-training process. It offers both cloud-hosted, software-as-a-service products and on-premise products, and its API allows customers to deploy trained models with a minimal amount of code.

“The applications of AI are limitless,” Huang said. “Our goal is to enable humans to do more of the fulfilling and high-value work they love and less of the tedious tasks that consume their time and resources.”

R2’s clients kick things off by defining the problem, or problems, they hope to solve with AI. Next, they collate the relevant data — with the help of R2’s team, if necessary — and upload it to the R2 Learn cloud platform, after which it’s inspected and preprocessed to prevent bias, skew, or overfitting in the resulting model. R2’s automated toolset then trains several models and identifies the optimal one by testing it with a portion of the training data. Finally, it’s delivered to the customer for validation.

Most consultations take about 48 hours from start to finish, Huang said, and the results so far are encouraging.

A recent customer recruited R2 to develop a model that would predict patients’ risk of type 2 diabetes, Huang told VentureBeat. The system took just 20 minutes to train and far outperformed a human panel of experts, achieving 89 percent diagnostic accuracy compared to scientists’ 78 percent.

Another client — a small hedge fund — used R2 Learn to generate a rolling stock prediction algorithm that trains on new data every day, improving its accuracy over time.

“[We] are developing […] solutions to empower businesses of all sizes to turn big data or human expertise into AI solutions and enable AI to have humanlike creativity,” Huang explained. “All of this will be done without a significant investment of time, data scientists, or big data resources.”

R2 Learn will launch publicly this fall ahead of R2 Suite, R2’s full-stack AI development solution. The company, which has offices in Shanghai and Hangzhou, has 26 employees. It’s raised $3.1 million in seed funding to date.