metamorworks - stock.adobe.com
The technology division of L'Oréal, a longtime Google enterprise customer, subscribes to Google Cloud's Vertex AI platform to speed up the production of its AI models for cosmetic services.
So far, it has worked well, said Jeff Houghton, chief operating officer of L'Oréal's ModiFace, which develops augmented reality and AI digital services for the beauty industry.
Google's Vertex platform, formally unveiled on May 18 at the Google I/O virtual conference, enables customers to accelerate the development, deployment and maintenance of machine learning models by unifying Google Cloud Services, machine learning systems and machine learning operations (MLOps) under one user interface and API.
Google claims Vertex cuts 80% of the code customers typically need to generate before getting their in-house AI algorithms to work properly on local hardware. While there is no agreed-upon benchmark to prove those results, the benefit of using Google's TPU v4 processor pods and upgraded AI software platform is undeniable, Houghton said.
"When we were training algorithms before, we would have to run millions of test images," Houghton said. "Now we can rely on the Vertex technology stack to do the heavy lifting. Vertex has the computing power to figure out complex problems. It can do billions of iterations and Vertex comes up with the best algorithms."
That represents a major savings in production time and effort, said Houghton, whose team of 70 developers includes nine AI specialists. ModiFace develops algorithms that power digital services for L'Oréal and some of its other subsidiaries, including Maybelline, Armani and Lancôme.
What this means, ultimately, is that ModiFace, a beta tester of Vertex, can focus on innovation rather than time-consuming model production, according to Houghton. Using Vertex, for instance, ModiFace recently launched a service that enables customers to digitally "try on" nails in as realistically a way as possible.
That is, customers can use ModiFace's augmented reality to visually apply various faux nail sizes, shapes and colors to an on-screen hand whose nails are realistic-looking.
"Tracking hands research was there before but with the evolution of AI, we can understand the orientation of the nail much better," Houghton said.
Vertex enables enterprise customers to subscribe to the AI tools and computing power developed in-house by Google Research to power Google search, including computer vision and natural language technology, according to Google.
That is perhaps the core value of Vertex -- it extends Google's prize technologies to enables developers to use Google's neural network search architecture in Vertex AI, Google execs said during the May conference.
Accessing Google's TPUv4 chip pods and AI software is valuable to customers but priced fairly as a service, Houghton said.
"Vertex helps us come up with the best algorithm … and once the heavy lifting is done, we switch off," Houghton said. "We use it on a subscription model and the amount of computing power needed changes. The cost will scale with our needs."
Jeff HoughtonCOO, ModiFace
In addition, with Vertex AI's new Vizier tool, Google is extending its MLOps tools to enterprise customers to accelerate AI model development, which speeds experimentation of algorithms, as well as the Vertex Feature Store for reusing ML features. Google's Vertex Experiments is the feature that accelerates deployment of AI models into production.
ModiFace launched its first augmented reality service in 2010 and has evolved in step with Google's AI toolkit. L'Oréal's search, for instance, is powered by ModiFace renderings.
But Vertex takes AI to a new level for enterprise customers, Houghton added.
Houghton said he plans to experiment with more of the Google MLOps tools available to his team, but the most value Vertex offers is time saving in more mundane aspects of AI model development, and access to the tech giant's TPU v4 machines.
As Google AI advances, ModiFace can take its AI development to a much higher level.
"It means we can innovate more," Houghton said. "Now we can train AI models to do more things with facial characteristics," Houghton said. "We can come up with new services that revolutionize the beauty tech space."