Novera files a patent for a new system of skin diagnosis AI “Viewty” that is more fairer to skin
Since they officially released cosmetics recommendation app Viewty at the end of 2019, with its core technology of skin diagnosis AI, Japanese company Novera Inc. has also been providing a wide range of solutions for brands and other companies. Novera has overcome the issues that arose while trying to improve the accuracy of their skin diagnosis AI and has come up with a radical new solution — a patent for which was filed in March this year.
Novera, a company focused on developing AI for skin diagnoses, possesses a data set of roughly 30,000 patterns just for “skin quality types” and is an expert in skin-related data science. One year and 4 months into the release of their recommendation app Viewty, the skin data that they’ve gathered using their in-house methodologies currently tally over 40,000 entries and over ten companies are making use of it in the form of an API (Application Programming Interface).
A major feature of Novera’s skin diagnosis AI is how it incorporates the knowledge of over 200 beauty specialists, including currently active beauty advisors. While many of the skin measuring AI programs developed by Western companies use dermatological knowledge for an approach based more on medical treatment and health care, Novera instead concentrates on everyday beauty and is trying to position their product as “skin diagnosis AI for everyday use”.
Novera CEO Kunitaka Endo says that a big change is occurring in the field of skin diagnosis as a whole: “In stores, the skin measuring methods that use custom devices and that require direct contact with the customer is difficult in terms of preventing viral infection. So what we’re looking at is contactless digital skin measuring where a photo is taken via a smartphone camera.”
When brands select a type of skin diagnosis AI to implement, the decisive factors include the conversion rate, its effect on gaining customers and on advertising, and its specific impact. While Endo acknowledges these points, he believes that the first step in promulgating skin diagnosis AI lies in lowering the threshold to usage.
“At Novera, in order to lower the threshold to implementation, we’re keeping the initial implementation fees low and offering an ample selection of trial services. Also, we’re thinking about not just grouping our products, but partnering with brands and media outlets to build up a variety of business models.”
Patenting skin diagnosis AI development technology that combines advanced expertise and generalization ability
On top of this, Novera is currently continuing to brush up on technical aspects such as the accuracy of their diagnosis results. One such project is called a “learning method for improving the accuracy of skin AI” and they’ve just recently filed a patent for it.
Endo explains: “While we were allowing the skin diagnosis AI to learn the knowledge of 200 professional beauty advisors and the accuracy of the diagnoses continued to improve, a problem arose. This was a phenomenon where the diagnosis results were extremely biased towards a particular topic, such as “sensitive skin”. Although the data taught to the AI isn’t biased, the output showed a bias for some reason. One possible reason for this was that the beauty advisors couldn’t completely put their knowledge into words. Our analysis found that there was a gap between what they wanted to say and their ability of expression and that this led to biased diagnosis results. After trying to solve this through a trial-and-error approach, the result was the filing of a patent for a learning method for improving the accuracy of skin AI.”
Novera’s AI supervisor Daiki Morotomi came up with a trailblazing solution for this issue.
Morotomi: “A phenomenon similar to “overfitting” — to use a technical term — had occurred. This is a phenomenon where the AI is taught certain training data however it can’t apply to unknown data. After we conducted a detailed examination, it turned out not to be overfitting, but a similar phenomenon had occurred, which was causing an extreme bias. We tried dealing with it by adjusting the variability and parameters of the conventional structure and fixing the advisors’ data, however, it didn’t solve the problem.” Even though the AI delved solely into the specialists’ field of knowledge, it couldn’t handle more general knowledge, and so it comprehended things in extreme ways, and this also made it seem like it was making biased judgments.
The solution for this was their “learning method for improving the accuracy of skin AI”. While concrete details can’t currently be divulged due to the patent being under evaluation, the method has yet to be announced in academic papers around the world and Novera has hinted with confidence that, in terms of the field of skin diagnosis AI for everyday skincare, it’s a “technological development that other companies’ services cannot match”.
Additionally, this learning method can not only help improve the accuracy of skin diagnosis AI but can also be used to improve the generalization ability of other types of AI, such as AI that links skin diagnosis results with purchasing data. Henceforth, Novera plans to use this patent as a means to provide AI development models and engines to brands and other companies and in turn drive their business forward.
Says Endo: “If for example, you want to find insights regarding when an existing user switched to a different brand or product, there are going to be limitations with just text-based questionnaires or purchasing data. However, by combining highly accurate skin diagnosis AI with the data you currently have, we think you can find out in detail the relationship between product purchases and changes that are seen in users — for example, individual skin conditions — and environments — for example, seasonal influences. We want to continue to update our skin diagnosis AI solutions and at the same time search for ways of doing business in a variety of different directions.”
Text: Ching Li Tor
Original text (Japanese): Jonggi HA
