Google announces "Cloud AutoML Vision" to make it easy to build a custom machine learning model using its own data set

The Google Cloud AI team aiming for "democratization" of artificial intelligence (AI) can easily create a custom machine learning model for image recognition even for people without much specialized knowledge "Cloud AutoML VisionWe announced.

Cloud AutoML - Custom Machine Learning Models | Google Cloud Platform
https://cloud.google.com/automl/


Cloud AutoML: Making AI accessible to every business
https://blog.google/topics/google-cloud/cloud-automl-making-ai-accessible-every-business/

Based on Google's image recognition API, "Cloud AutoML Vision" announcement that let you learn favorite images and easily customize recognition function - Publickey
http://www.publickey1.jp/blog/18/googleapicloud_automl_vision.html

The number of people who can make sophisticated machine learning models is extremely limited and it takes time to build our own custom machine learning model, so it can not be said that "anyone can use it freely." The Google Cloud AI team, which is actively working to think that this gate should be wider, has provided the service "Google Cloud Machine Learning Engine" for some time, but even though he is still a skilled engineer in machine learning Was necessary.

However, as machine learning and artificial intelligence booms are occurring, the number of experienced engineers is in short supply. As a result, the Google Cloud AI team has announced the launch of "Cloud AutoML", a new service that allows engineers who are unskilled to build their own high-quality custom machine learning models by utilizing Google's advanced technology . The constructed machine learning model can be used via Google Cloud API.

At the same time as the first release of the corresponding API, it is possible to recognize the image according to each use by learning the image data prepared by the user "Cloud AutoML Vision"is. You can see how you learn and what you can do by looking at the introduction video published by Google.

Introducing Cloud AutoML - YouTube


Google has provided "Cloud Vision API" for image recognition to be used in combination with "Google Cloud Machine Learning Engine".


However, even if meteorologists try to detect cloud differences such as "cumulus clouds", "cumulonimbus", "stratum clouds" using this API for weather forecasting and flight plan creation ......


Only information such as "sky", "cloud", "daytime" is detected. This is because the Cloud Vision API outputs the results of labeling based on Google's pre-learned information and the types of clouds are not included in advance study content.


If you learn a cloud image of a large amount of climate that a meteorologist has, you should be able to recognize the difference of clouds. To that end, we had to build a machine learning model on our own.


Cloud AutoML Vision solves this problem to "simple" "safe" "flexible".


Even if you do not have an engineer familiar with machine learning, you can build a custom machine learning model by uploading your own dataset.


Cloud AutoML UI looks something like this. In the example of the previous cloud, first upload a picture of a lot of clouds. It seems that labeling can be done before uploading or after uploading.


The necessary conditions for learning are 2 to 100 labels, 20 to 100,000 labeled pictures, and at least 10 pictures for one label. When you are ready click "Train" button.


After learning, you will be able to see the analysis and statistical results.


Uploading photos of the cloud at the end, according to the contents of the previous learning, predicted "This is a high probability cirrus".


"Cloud AutoML" is currently in alpha version,We accept test participation participation.