Ivan Zhou

View Original

AutoML from Google Cloud — Free Up ML Resources to Focus on Harder Things

Our first Cloud AutoML release will be Cloud AutoML Vision , a service that makes it faster and easier to create custom ML models for image recognition. Its drag-and-drop interface lets you easily upload images, train and manage models, and then deploy those trained models directly on Google Cloud.

Google Cloud Platform announced the alpha launch of AutoML Vision this week. The service aims to help developers with no machine learning expertise to easily create a customized and well-tuned machine learning model for image recognition.

 

It is not the first system of its kind on the market: Clarif.ai, Microsoft’s Cognitive Services, and Salesforce Einstein Vision all provide similar services to the market.

It neither means Machine Learning engineers will be replaced: some repetitive work will be automated; however, those more complicated and domain-specific machine learning works still require a lot of machine learning experts and data engineers.

However, if the AutoML service proves to be effective and expands to other areas beyond vision, such natural language generation and anomaly detection, two groups will largely be benefited from it:

First, the Google Cloud team can provide more AI-driven services to their clients, with AutoML as building blocks to higher level customization work. This do not necessarily mean they will get higher profit margin, as other competitors like AWS and Azure may quickly release similar services. Instead, it serves as a toolbox for every Google Cloud consultant: the tool can be used to can speed up the repetitive, low-level work, so that the consultant can focus on the more difficult, domain-specific problems and quickly deliver customizing solutions.

Second, more importantly, those who have very limited access to the machine learning expertise and data resources, such as startups, University researchers (in non-CS and Engineering fields), and social ventures, can close their gaps with large organizations, and focus on solving problems in their domains. For example, psychology researchers can leverage the powerful image recognition service to detect facial changes of subjects in the video of experiments, without the need to hire a bunch of undergrad students to do so; more sociology researches like this one from Stanford can be carried out more easily.

 

Research from Stanford on Estimating the Demographic Makeup of Neighborhoods with Deep Learning

In short, AutoML and other similar services do have the potential to make AI more accessible to everyone, so that we can focus on much harder problems. We are far from automating the entire pipeline of Machine Learning development.

Research from Stanford on Estimating the Demographic Makeup of Neighborhoods with Deep Learning

In short, AutoML and other similar services do have the potential to make AI more accessible to everyone, so that we can focus on much harder problems. We are far from automating the entire pipeline of Machine Learning development.