Local neural net data processing is not far away to become reality
SO over the years the use of AI is increased and it changing the way we interact with out Mobile phones
AI or Deep learning need lots of power for running locally devices like Amazon echo using Alexa, perform all or part of the processing in the cloud.some applications cant run in cloud we need to run locally which also provides the additional security of data.
Then Embedded AI comes in the spotlight Which is directly integrated into devices . Embedded Ai needs massive amount of power to run locally and MediaTek is looking to make local neural net data processing a reality.
Mideatek is announced the Helio X20 MediaTek developed what it calls the Deep Learning SDK for developers. The SDK is optimized to work with the TensorFlow and Caffe deep learning frameworks.
Thus far, MediaTek has worked with undisclosed partners on developing solutions for image and voice recognition that leverage the heterogeneous computing capabilities of the Helio X family of mobile SoCs.
The helio X family has 10 CPU Cores, combined with GPU cores, Image Signal Processing (ISP) cores, and Digital Signal Processing (DSP) cores.Basically the goal is models should be able to continue learning and modifying.However, with so many compute resources on a single chip, on-device learning may not be far away.
Acording to Mediatek the Currently the targeted product is Smartphone but the platform could easily be leveraged for other applications, such as home entertainment or automotive infotainment