Article Directory
1 Introduction
First of all, I would like to thank CSDNfor providing a directional invitation ticket, which made me have the honor to participate Google开发者大会. Of course, this is also my first time to participate. Google开发者大会It is a developer conference that truly belongs to technical people. The content of the conference is all technical dry goods, and the content is quite rich.
When each big coffee shares technology, he will first introduce the current technology and the technical field to be overcome in the future. Of course, we will also use actual data and cases to show you how far the current technology has achieved. Before the end of each sharing, there will be a corresponding QR code for everyone to understand and experience related technologies.
2. Conference content
The conference lasted for two days (September 14th and 15th), and mainly included 13 contents. You can directly see the contents on the pictures, and choose the corresponding session according to the content you want to hear. Information about the conference can also be found online.
Because I am interested in the content of machine learning-tensorflow, I chose ticket No. 15. Here is also sorted out the tensorflow-related content mentioned by the big names in the next conference, and share it with you here.
3. Intensive Lectures on Tensorflow
What to share: Google provides developers with a comprehensive ecosystem of open source machine learning products.
Speaker: Wei Wei
tensorflowis 机器学习the most widely used module in China. Whether it is computer, medical education, etc., there are tensorflow figures. Google hopes that some non-professional fields can also use machine learning. In order to reduce the entry cost of machine learning, Google also Studies have been done to this end and some examples are provided. What impresses me the most is Magic Erase技术that it Pixel设备uses machine learning technology on the Internet to erase unwanted content from photos. Before we edited pictures, we used Meitu repair or PS technology, but the introduction of machine learning can be said to provide convenience. (Review the machine learning techniques used around us, for example, in the use of camera stickers, some are dog face recognition and cat face recognition.)
3.1 Machine learning process
For the process of machine learning, the speaker also gave a detailed introduction
- Data injection and preprocessing
- Model architecture definition and training
- model deployment
- Monitoring and Maintenance
3.1.1 Data injection and preprocessing
Source of data sets: There are many trained data sets on the Internet, such as https://url.com/ , or search in https://www.kaggle.com/ (some papers will also provide some data sets Source)
Data preprocessing refers to filling and normalizing the default values, outliers or large data values in the data set, and processing them into a data set that is easy for the model to process.
3.1.2 Model architecture definition and training
Because our data sets may be pictures, text, audio, etc., we choose different model types for training for different data sets.
3.1.3 Model Deployment
After our model is trained, embed it into applications, such as Flutter applications, web applications, mobile applications, etc.
3.1.4 Monitoring and maintenance
Google has TensorFlow Liteintegrated it into Google Services, https://tensorflow.google.cn/lite/guide?hl=zh-cn , TensorFlow Lite is a set of tools that can help developers in mobile devices, embedded devices and loT devices Run the model for on-device machine learning.
4. Extra chapter
Google开发者大会It is a developer conference that truly belongs to technical people. The content of the conference is all technical dry goods, and the content is quite rich. And the whole conference is free. If you come from other places, they will reimburse you for your accommodation and bus tickets. When the exhibition area is open, there will be free small gifts, and you can experience the products and do tasks, and there will be corresponding benefits.
