Google Cloud Enters China Through Capital Online? Can AutoML Be Used?
By Super Neuro
Before I could reply to Google Search, I received the news that Google Cloud was going to enter China.
Could it be that AutoML can really be used conveniently?Google Cloud is coming?
This afternoon, the official WeChat public account of Capital Online released 「Google Cloud will enter China through Capital Online」news.
Since a few months ago, when Google executives frequently visited China, established an artificial intelligence institute, and stated that they had never given up on the Chinese market, many people have been looking forward to the return of Google-related products.
Although Capital Online deleted the article shortly after it was published, it privately stated "Capital Online only has an overseas agency agreement, not China as mentioned in the article."
However, the tone of this article is affirmative and the wording is clear. As for the reasons for deleting the article and denying it, they are still worth pondering.
Although it is not certain that Google Cloud will be able to enter China smoothly, and it may take some time for AutoML to be conveniently used, there are still many other similar services.Besides AutoML? Although there are many ready-made machine learning model frameworks that can be used directly to reduce the cost of building AI models, and many companies have also open-sourced their own ML tools.
However, these frameworks cannot achieve the same standards as the others. They are difficult to interoperate with each other, and there is no effective solution to allow ML frameworks to be connected to any application. If enterprises need to use multiple ML model frameworks at the same time, they still need to complete a lot of engineering work themselves.
To solve this problem, Salesforce and Oracle have each developed tools that can connect these open source ML model frameworks in different applications.
Previously, although there were solutions to the above problems on the market, such as establishing interconnected APIs, such as Python-JSON API, this solution would lose the performance of the ML model framework while achieving docking.
Even if the ML model temporarily solves the docking problem through API, enterprises still need to build a dedicated model server to commercialize it. Building a server is not only expensive but also quite complicated. For example, it may take several days to build a GPU version of TensorFlow-service.
Therefore, the main problems that currently hinder the widespread use of ML model frameworks are the lack of standardized application programming interfaces and the high threshold for building model servers.
As a result, frameworks that can connect these ML models in different applications have emerged. TransmogrifAI: ML framework in the cloud TransmogrifAI is an ML framework based on the Apache Spark engine that can perform feature engineering, feature selection, and model training. It can also integrate existing ML models to match the most cost-effective ML model for any application, without requiring enterprises to create a separate model server.
The framework was developed by Salesforce. This veteran enterprise service company is not satisfied with the status quo and has also quickly caught up with the AI trend.
Recently, they open-sourced TransmogrifAI.
The AI platform Einstein built by the company is one of the largest machine learning projects in the industry, with advanced machine learning algorithms, natural language processing, and intelligent data mining capabilities. The platform can now connect to most ML models on the market and simplify these models.
What enables Einstein to do all this is TransmogrifAI.
TransmogrifAI draws on the principles of AutoML to simplify the machine learning operation process and improve developer efficiency. TransmogrifAI has four basic principles: modularity, compilation security, transparency, and automation.
These four principles have been transformed into a simple programming model, allowing engineers to complete tasks such as data organization, feature engineering, and model selection by writing only a few lines of code. GraphPipe GraphPipe can serve machine learning models made with popular frameworks such as TensorFlow, MXNet, Caffe 2, and PyTorch in the cloud.
The purpose is to lower the threshold for using ML models so that AI models can be used in mobile applications, IoT devices, and the Web.
It is an efficient network protocol that simplifies and standardizes the transmission of machine learning data between remote/processes, allowing users to flexibly choose the appropriate machine learning model under the existing framework. This means that developers do not have to build APIs specifically to connect to AI models, nor do they have to bother to study which ML framework can better create AI models.
In addition, GraphPipe has launched a series of open source tools for AI developers for popular frameworks such as TensorFlow.
Currently, TransmogrifAI and GraphPipe are available for free on GitHub.Super Neuro Encyclopedia
TransmogrifAI Architecture:
https://www.colabug.com/4152476.htmlMachine Learning - Feature Processing:https://blog.csdn.net/kanbuqinghuanyizhang/article/details/78993386
Introducing GraphPipe:
https://blogs.oracle.com/developers/introducing-graphpipeOracle open source Graphpipe:
Salesforce open-sources TransmogrifAI:
Feature Reasoning: