Google hosted Think With Google in Beijing on February 6. At the meeting, Li Jia, head of research and development at Google Cloud AI, delivered a keynote presentation focusing on Google Cloud AutoML, which was released two weeks ago. However, most notably, The first title on the conference has quietly become president of Google's AI China Center. After the conference, Dr. Li Jia accepted an interview with NetEase Intelligent about the work of Google's AI China Center and the future development of Google Cloud AutoML.
Figure: Google China AG president Li Jia
After more than two months after the establishment of the Google AI China Center, Li Jia appeared as his new president.
In November 2016, Li Jia left from Snapchat and joined his fellow teacher at Stanford University, Li Feifei, to become Google's head of cloud AI research and development. In December 2017, Li Feifei announced at the Google Developers Conference the establishment of the Google AI China Center The two men jointly led the news, and now Li Jia has become the executor and director of the new department.
As the leader of Google's AI China Center, Li Jia made a presentation in the recruitment information, vigorously recruited machine learning researchers and engineers.In the interview, Li Jia also said that Google AI China Center will focus on basic research, and at present The main job is to gather outstanding AI talent.
In addition, because Li Jia and Li Feifei are both Google Cloud executives recently released an important development tool Cloud AutoML, so we have reason to believe that Google AI China Center in addition to the early tasks in addition to recruitment Will vigorously promote Google Cloud AI concepts and a series of tools.Li Jia said in his speech, Google Cloud AI's goal is to promote the positive interaction between AI research and product development.And the recently released Cloud AutoML become an important Google Universal AI tool, which gives AI products a low threshold.
It is reported that Cloud AutoML released two weeks ago, the registered users has more than 10,000, and the user's background is also quite extensive.In addition to the retail industry, the medical industry, users, there are many special cases.For example, the London Zoological Society 'camera' Logging 'Keep track of animals around the world and auto-tag images with Cloud AutoML Vision, saving you a lot of workforce tagging, as well as some environmental agencies using this tool to detect junk and more.
Li Jia revealed in an interview that Cloud AutoML has been developed for some time now and that Google Cloud's team has done a lot of work before, one of the most difficult and crucial issues being the combination of technology and products. In research and development, we also need to consider the power and resource needs. Finally, Cloud AutoML integrates Transfer Learning and Learning to Learning, both of which complement each other's strengths, said Li Jia.
Cloud AutoML is really a handy tool from an AI application perspective.If an enterprise is customizing an AI solution, it's time to define an algorithm that collects data, typically months or years to complete.And Cloud AutoML automatically generates Model, product cycle is greatly reduced, the cost of business will be reduced.Moreover, Google also claimed that its classification in the picture than the expert design model works better.
After Cloud AutoML Vision (image), Google will also launch Cloud AutoML in the areas of voice, conversation and natural language processing.
But does low threshold Cloud AutoML grab the AI developer's job? Li Jia does not think so, she said. 'There are only 1 million AI professionals in the world who can build their own algorithms, but 21 million developers around the world can use it AutoML such products. 'Li Jia said that Cloud AutoML is actually a product for developers to create, it lowered the threshold for more developers to use.If there are more people using AI, these developers The chances of becoming more, then there will be more to understand the industry, people who understand the product their expertise to the extreme.
For the development of AI users, Li Jia will be divided into four categories:
The first is an AI researcher who can design and develop tools such as TensorFlow.
The second is the API users, these users do not have the data to do training, they need to interface, not to develop their own model.
The third type of user wants to customize their own model. They do not need to recruit AI talent themselves, do not need data labels, and even customize their own model with Cloud AutoML without the need for tuning.
The fourth user is mostly traditional industries, they do not know how to use AI, do not know how to do this time Google will work closely with customers, teach them how to use machine learning.
When asked how to keep Cloud AI competitive in technology giants, Li Jia said Google is an AI First company and that all of Google's current products are Powered by AI, AI is a great fit with Cloud, behind Google's iteration of products and understanding of AI for more than a decade, and we want to share that experience with users so they can create more products .
'AI itself has no boundaries, and we hope that AI technology will be used in the world in the future,' Li Jia said.