[ad_1]
The highlight of these accelerators (of any brand) is that they can perform more operations in less time than any other traditional circuit. 19659002] The novelty is that now Google has launched a new version of these chips, called TPU Edge, which will perform machine learning tasks on IoT (Internet of Things) devices.
The process of automatic learning involves, in broad strokes, two steps The first is in learning the algorithm with data. The TPU, launched two years ago, is dedicated to this task.
The new Edge TPUs are designed for the second stage of machine learning, which is the inference, that is, the interpretation of information ( it is the execution of algorithm). This is what allows, for example, to recognize objects in a photo.
The main advantage of the inference made in the device – instead of having to send the data to the Internet to be badyzed – is that it is earned in security and the speed in the processes.
As to their application, these new chips are designed for use in factories where, for example, processes such as quality control could be automated
. This is the only company that has this type of products. Qualcomm, Mediatek and ARM, among others, also manufacture artificial intelligence accelerators.
The difference is that Google offers a complete solution, as highlighted during the event Google Cloud Next 2018.
Anything can happen in one place: you can store data in Google's cloud, form the algorithms through the TPU, and then do the inference with the new TPU Edge
"The TPU Edge are designed to complement our offer of TPU in the cloud, to accelerate the machine learning in the cloud, and then produce inferences faster, sensors become more than just data collectors, they make intelligent decisions in real time, "says Injong Rhee, vice president of IoT Google Cloud in a statement
[ad_2]
Source link
