Artificial intelligence has been detonated by Google in the past two years. So the startup companies in the field have mushroomed. Some of them are dedicated chips, some are algorithms, some are robots... What exactly is artificial intelligence? Is it a machine that replaces humans to complete repetitive production line labor? This understanding is not too simple. Artificial intelligence is definitely not just for the machine to imitate, and thus to replace repetitive work. The real artificial intelligence requires the machine to be trained and learn to approach or exceed the human intelligent behavior.
When artificial intelligence training models, data is the most important factor. Only through big data training can the training model be closer to the actual needs. For example, the automatic driving model training is basically based on the self-driving car on-road test to collect data training models. On the road, the car only collects more possible scenes to ensure that the trained model is closer to the actual road conditions. There are two general problems encountered in artificial intelligence training: First, how much data is not enough to train the model? Second, how can data be quickly marked?
Zhao Zhihong, MathWorks Product Marketing Manager
Migration learning: training data through models to train neural networks If there is no data to do artificial intelligence? This is the confusion of many startup companies. In the author's concept, the normal AI development process is to collect data first and then train the model. At the recent Matlab conference, Mr. Zhao Zhihong, MathWorks product marketing manager, gave a different answer. He said, “Data, output, and models are a step in the entire AI development process. If the developer wants to do artificial intelligence, there is no Too much data can be migrated to learn.†What is migration learning? Do not prevent understanding from actual examples. For example, artificial intelligence predicts the failure of a wind turbine. Users cannot wait until a large amount of fault data is collected before predicting. This does not meet the goal of equipment maintenance. Instead, a Simulink model is used to build a wind turbine generator model. By correcting the model so that it is very close to the wind turbine and then generating fault data through the model, it can train machine learning or deep learning neural networks.
Mr. Zhao Zhihong said, “The advantage of model-based design is that it builds a model before it actually makes a product. This model is very close to the actual product, whether it's results from digital computing, behavioral calculations or real results. It's very close.Many engineers have built this model when designing the product.Adding fault conditions to this model is very easy to generate fault data, which is much easier than the fault that occurs on the actual equipment. Therefore, it can be faulted. Forecast and maintenance."
Labeling big data through deep learning With the large number of sensors used, the amount of data has entered the era of prehistoric times, and it has been difficult to characterize artificial intelligence. At this time, deep learning can be used for marking. Mr. Zhao Zhihong explained that “the feature of deep learning is that there is no need to manually find the feature values ​​manually. The system can automatically extract feature values ​​from the data. There are many tools and functions for automatic labeling in MATLAB. LiDAR 3D point cloud technology can be used for each point. Labeling, clustering this point together into a target model, and then distinguishing the physical representation of the target, some users have developed using MATLAB tools, and the famous auto parts company AUTOLIV is using this method. Automatic data annotation."
Taking puffed food smart testing as an example, researchers can extract features when users bite foods and measure the crispness of foods with biting sound and bite force. With these two characteristics, a machine learning classifier needs to be developed. While MATLAB provides a classifier learner. Developers do not need to go through all the algorithms of the various classifiers one by one by using this tool. Use the APP provided by MATLAB to try all algorithms at once. After running the APP, the user selects the data and the classifier to be trained, and then trains. During the training process, users can see the overall results of each classifier, choose the one with the highest accuracy, and then conduct more investigations and studies. If the user has not studied AI, he can use the APP provided by MATLAB to learn and try all algorithms for machine learning.
Is it difficult to do deep learning in MATLAB? In the words of Mr. Zhao Zhihong, only five lines of MATLAB code need to be written so that users can build a network that can identify food. The author has not been in contact with MATLAB for many years, so there is no research, engineers and friends can try if necessary.
How to collaborate with open source code? Now that open source is very hot on the market, engineers can quickly obtain the required code in the open source community. There are also many open source resources in the field of artificial intelligence. Compared with open source code, MATLAB as a closed resource may be slower to update, and it is The charging model, engineers are also worried about the integration of the two issues. When the author asked this question, Mr. Zhao Zhihong explained that “There are many development teams behind open source code, they are willing to develop, and we also have our own development team. We do not define open source code and MATLAB as competitive relationships. Each of us has its own characteristics and strengths, and sometimes we can share it with open source code.There are some open source code that can be run in MATLAB or directly called, and we are working together to contribute to the progress of science and technology. Although we are paid software, we provide not only Tools, there are many services."
"When users face problems with MATLAB, we have a strong technical support team behind them to help solve these problems. In addition, our tools not only produce correct results for a certain function, we also consider this function very Easy to use, for example for AI applications, we make the interface well-suited for this area of ​​expertise, but also allow users to feel familiar with the familiar MATLAB or Simulink interface.This is our advantage if the user adds new in MATLAB or Simulink Functionality, learning threshold is not very high, easy to design. In addition to providing tools, we also provide users with a lot of convenience, which is something that can not be obtained from open source tools.†Mr. Zhao Zhihong added.
Fork Type Connecting Terminals
Fork Type Connecting Terminals,Terminals,Connecting Terminals
Taixing Longyi Terminals Co.,Ltd. , https://www.longyicopperterminals.com