It’s a well-known fact that the use of deep learning has affected all sectors. It has taken many recognition tasks and entire computer world by storm. MOST of the difficult recognition problems have gained over the past few years which are astonishing.
In the recent times, we have seen a large improvement in the accuracy of recognition systems due to the use of Deep learning approaches. However, the main challenges for deep learning are that they require the specific type of computer which is quite different from what we use on general purpose and also that they need a large volume of data for training. There are few companies who have used special architecture to address the particular type of parallel computer required for DNNs. These approaches have enhanced the capabilities of cloud computing a lot.
Microsoft has decided to address the same issue. HoloLens contains a custom multiprocessor called Holographic Processing Unit or HPU. This is responsible for processing the information coming different sensors including, Microsoft’s custom time of flight depth sensor, infrared camera, head tracking cameras. It is the HPU that makes the HoloLens the world’s only fully self-contained holographic computer.
Microsoft’s executive vice president of Artificial Intelligence Group announced that the second version of HPU, which is currently under development, will use an AI co-processor to implement DNNs. This chip will support a wide variety of layer types which will be programmed by Microsoft. The AI coprocessor is designed for the next version of HoloLens and will work on the HoloLens battery.
This is one of the major steps toward developing mixed reality headsets. Microsoft has already indicated that they are more dedicated toward working with the mixed reality rather than virtual reality or augmented reality. Microsoft earlier said that the company considers artificial Intelligence and mixed reality as the future of computing and this could be one of the major steps toward achieving that.
Video & Image Source: Microsoft Research