HoloLens, Microsoft’s self-contained augmented reality headset, is set to receive an artificial intelligence (AI) upgrade.
The Redmond, Wash. technology giant is working on an AI co-processor for the next HoloLens, the company revealed this week. Currently, HoloLens packs a custom chip dubbed the Holographic Processing Unit (HPU), which enables the device to map out a user’s environment and anchor virtual objects to those surroundings, among other tasks required to turn sensor data into convincing mixed-reality experiences.
The next HPU will feature an AI co-processor with deep neural network capabilities, announced Harry Shum, executive vice president of the Artificial Intelligence and Research Group at Microsoft, during this week’s Computer Vision and Pattern Recognition conference in Hawaii. The custom silicon will allow HoloLens to perform real-time, AI-enabled analysis on sensor data gathered by the device independently of another system.
During the event, Microsoft offered an early peek at the technology by demonstrating hand segmentation techniques used in gesture recognitions systems. “The chip supports a wide variety of layer types, fully programmable by us,” blogged Shum’s colleague, Marc Pollefeys, director of science at Microsoft HoloLens. “The AI coprocessor is designed to work in the next version of HoloLens, running continuously off the HoloLens battery.”
Microsoft seems to be living up to its promise to bring AI to all.
For Windows and Linux developers, the company recently released Security Risk Detection—formerly Project Springfield—which is a cloud-based fuzz testing service that uses AI to find flaws in software that may cause a crash or let attackers compromise a system. Earlier this month, the company released an iOS app called Seeing AI for the visually impaired that provides an audio description of a user’s surroundings based on the information captured by an iPhone’s or iPad’s camera.
The competition is also bullish about the market for AI solutions and like Microsoft is developing AI chips of their own to help spur adoption.
Gartner recently predicted that AI will somehow figure into practically all of the new software products and services released by 2020. By that same hear, it is expected that nearly a third of CIOs (30 percent) will list AI as one of their top five investment priorities.
In May, Google announced its second-generation Tensor Processing Units (TPUs), which the search and online advertising giant is incorporating into Google Cloud. Initially available through the company’s Compute Engine cloud virtual machine service, the new TPUs speed up the process of building and training machine learning models.
At Google, the chips are already paying off, allowing the company to accelerate AI workloads using “TPU pods” that contain 64 TPUs and provide up to 11.5 petaflops of processing power.
“One of our new large-scale translation models used to take a full day to train on 32 of the best commercially-available GPUs [graphics processing units]—now it trains to the same accuracy in an afternoon using just one eighth of a TPU pod,” reported Google staffers Jeff Dean and Urs Hölzle, in a May 17 announcement.