This afternoon, at the GTC China 2016 (GPU Technology Conference) held in Beijing, the Nvidia Institute of Deep Learning introduced a simple introduction to image recognition in CNTK.
First of all, let's learn about CNTK.
CNTK (Computational Network Toolkit) is an open source computing network toolkit developed by Microsoft Research. It describes a series of calculation steps through a directed graph and is also a unified deep learning neural network toolkit.
The overall architecture of the CNTK is shown in the above diagram. Here we can simply understand it in several parts, namely the upper-middle-down and the left-right-two-understand logic. Let's talk about the upper, middle, and lower parts , and the bottom one contains data that can be understood based on the IDataReader layer .
Imagine training an apprentice in a pawn shop to verify the authenticity of the antiques. You need to get some real antiques to let him understand. This is the process of learning and the process of inputting initial data.
The result of the learning is that the second layer contains CN
It can be understood as a pawn apprentice who realizes the knowledge of the brain through genuine and fake antiques.
The top IExecutionEngine
In fact, it can be understood as the personal behavior of this pawn shop apprentice. The customer took an antique and allowed him to identify his or her true and false. This is the top level role or meaning.
Secondly, the left, middle, and right parts , the leftmost are the input raw data. Here we introduce the CN Description, which is the description of the network.
Can be understood as guidance given by the master of apprentice pawn or antique identification textbook
The Features in Features & Labels are samples or antiques. Labels can understand that the samples on the antique samples are genuine and fake labels.
The middle is that ICNBuilder is the driving force of learning, and IDataReader is the drive to carefully observe and test the behavior of what you have learned. The right-most Ileaner is a learning method. Before CN explained that it was the result of learning, IExecutionEngine is the act of finally taking the real thing to test.
It can be said that the overall architecture of CNTK describes the learning process from entry to mastery for a pawnshop apprentice . Train yourself from the sample, form your own abilities, and use what you learn.
According to the researcher's description, the image recognition speed in the CNTK Toolkit is much more popular among developers than the other four current mainstream computing toolkits due to better interaction capabilities.
In the latest release of CNTK 1.7 from Microsoft, CNTK has supported Nvidia's latest deep neural network database, cuDNN5.1 .
In fact, in the artificial intelligence research of Facebook, Baidu and other giants, Nvidia's GPU is often used.
Facebook's Big Sur server is designed around a high-power processor, the GPU, originally developed for image processing. A Big Sur server is configured for every 8 GPUs. Facebook uses NVIDIA's GPUs that excel at image recognition.
Baidu will apply Nvidia's GPUs to its unmanned system's on-board computers. This computer can solve a major problem faced by automotive manufacturers when integrating driverless technology: how to integrate artificial intelligence into a compact computer to help the car make decisions. In addition to providing powerful performance support for unmanned and complex calculations for deep learning, it can also drive multiple high-resolution display devices for use in dashboards, map navigation, and car dashboards in smart cars.
Almost all giants are relying on powerful graphics processing units (GPUs) for deep learning applications to handle more complex algorithms that can improve artificial intelligence and recognition capabilities. Microsoft is not far behind in this regard.
Poe Adapter,Ubiquiti Poe Adapter,Carrier Poe Adapter,Poe Power Adapter
Guang Er Zhong(Zhaoqing)Electronics Co., Ltd , https://www.geztransformer.com