![<?echo $_SERVER['SERVER_NAME'];?>](/template/twentyseventeen/skin/images/header.jpg)
AI artificial intelligence is currently the hottest industry issue. However, when people focus on technologies such as deep learning algorithms and cloud data centers, the actual AI applications of terminal devices are still being paid for, and the industry is still struggling to start importing AI applications. In order to accelerate the popularization of AI in terminal devices, Intel has introduced two computer vision software development tools (SDKs) that can quickly bring visual recognition deep learning applications to battery-powered devices and existing x86-based PC/IPC products. It will be able to drive a variety of innovative applications and enable more players to benefit from this wave of AI revolution.
Zheng Zhicheng, an associate of Intel Business Marketing Group Enterprise Solutions, said that AI technology has been in development for nearly 60 years. The rise of this wave of enthusiasm is mainly due to the breakthrough of deep learning algorithms and the improvement of big data and computing power. The processing cost of the data is greatly reduced. In addition, the ready and open framework of Caffe, TensorFlow, CNTK, MXNet and other frameworks is also an important key, so that more people can use this technology.
"However, looking at the current state of the industry, there are many people discussing Caffe and TensorFlow, but the actual application is only in its infancy. This is because the lack of relevant tools can actually put the models trained on these frameworks into practice. He stressed.
Therefore, in addition to a variety of chips, Intel has also developed a variety of libraries and platforms to enable these frameworks to achieve better performance. Among them, Intel provides two important SDKs, one is the Computer Vision CV SDK (Intel Computer Vision SDK), and the other is the Intel Movidius MDK. As an interface between Intel hardware chips and deep learning applications, it supports various framework training. The model can therefore quickly bring the inference function to the gateway or terminal device.
With a minimum power of 1 watt, the Movidius processor can be used in battery powered devices. For example, DJI-Spark is based on Intel's Movidius Myriad 2 vision processor and a well-trained model on the industry's well-known framework to achieve obstacle avoidance, face and gesture recognition. In addition, the new Google Clips mini AI camera also includes the Intel Movidius Myriad 2 vision processor, which can perform real-time AI and machine learning operations directly on the camera terminal, improving image recognition and processing performance, and capturing life at any time while offline. The wonderful moments in the middle and edit the image.
For devices that require continuous power supply, the CV SDK can be used to actually apply the trained model to the application. For example, Intel and Amazon's DeepLens is the first programmable deep learning wireless camera with Intel Atom X5 processor, Intel deep learning software tools, and Intel Deep Neural Network Compute Library, which can run computer vision models in real time on camera terminals. Support AI model training and import in the cloud environment, which can reduce costs and real-time response, helping developers to design more innovative applications of AI and machine learning.
Zheng Zhicheng said that the significance of the CV SDK is that it can support a variety of processors after the Skylake architecture, including PenTIum and FPGA accelerators. That is to say, the existing wide range of personal computers, industrial computers, cameras and other products, without the need for special deep learning chips, can transform the cloud-trained model into practical use, so that the AI ​​visual identification application can be implemented. On a variety of devices, and not a distant talk.
After adopting the adoption of index manufacturers such as Google and Amazon this year, Intel is more confident in promoting the AI ​​edge computing market. In addition, in terms of speech recognition, Intel has also developed a GNA chip (Gaussian mixture model and neural network accelerator) and SDK, which can also bring the most popular intelligent voice functions to very low power devices with low power consumption. Currently, it is seeking the adoption of major customers.
Zheng Zhicheng said that bringing the AI ​​inference function to the terminal device is bound to drive the trend and become the focus of a new wave of AI. This will promote the popularization of AI in all walks of life, especially for Taiwanese companies. With the opening of various frameworks, the industry no longer has to invest huge amounts of cost and time to develop their own algorithms and build them from scratch. Complex hardware architecture to train neural networks. Taiwanese companies should step up their efforts to consider how to use these existing resources and tools to accelerate the introduction of actual AI applications.
Taking industrial computers as an example, combined with visual recognition, it can support automatic optical inspection (AOI) operations without the need to purchase special equipment to improve its manufacturing quality. Therefore, even small and medium-sized businesses can afford AI applications at the price of PC-grade industrial computers, and can really benefit. This is the strategy that Taiwanese companies should focus on, and strengthen the practical application of AI to enhance industrial competitiveness.
Intel, which has recently actively cultivated the AI ​​market, has built a complete AI computing platform from the cloud to the gateway and terminal devices. Zheng Zhicheng pointed out that this wave of AI rejuvenation mainly uses neural network algorithms to achieve a major breakthrough in training and inference. However, in terms of deep learning training, neural networks are completely different from the traditional concept of traditional IT data. Once a new identification object or situation needs to be added, it takes a long time to re-train to add new recognition capabilities.
Therefore, he believes that there will be many evolutions in the future of cloud technology to shorten training time and make AI applications more efficient. For Intel's layout, in the cloud data center, Xeon deployment will still be strengthened, and the Nervana platform will be used as a training accelerator; as for FPGA, it can be widely used in the cloud as an inference acceleration; , will support a wider range of Framework and topology through the SDK to open up new applications.
Zheng Zhicheng concluded that AI deep learning neural network technology actually covers a large scope, in addition to the development of algorithms, including data collection and labeling, analysis and pre-processing, etc. still need the CPU to deal with. As a leading chip industry, Intel is committed to finding opportunities and strengthening research and development cooperation with academia to promote innovation and implementation of AI applications at all levels.
Servo Motor Replacement Product Features:
1. Small footprint, high precision, high response, bus control, simple wiring, rich alarm functions, and lower cost;
2. All-copper coil, heat dissipation and waterproof; equipped with instantaneous power-off function to reduce downtime loss;
3. Customized motion control scheme: source manufacturers supply, strict quality inspection and guarantee; sufficient supply, short delivery time;
4. The speed can be adjusted arbitrarily within the rated speed according to the requirements of the computer program.
5. Optimize the electromagnetic scheme to improve the efficiency by more than 96%. Using superconducting materials, the temperature rise is reduced by 10%
6. The new movement has successfully realized the miniaturization of the motor.
Servo Motor Replacement,Robot Servo Motor,Digital Servo Motor,Perfect Pass Servo Motor Replacement
Kassel Machinery (zhejiang) Co., Ltd. , https://www.kasselservo.com