Cloud AutoML: For developers and enterprises first starting out on their ML adventures
Google is expanding its machine learning offerings with the all new Cloud AutoML. This service facilitates the use of machine learning models for developers and enterprises first starting out on their machine learning adventures. First up: image recognition!
Google’s been busy — it has expanded its range of machine learning options yet again. Google’s Cloud AI is a whole suite of services with pre-trained models as well as a DIY option. Highly accurate, easy to use, and scalable, it’s useful for businesses looking to get into machine learning without having a whole lot of expertise.
Currently, Google Cloud AI offers a number of useful services including the Google Cloud Machine Learning Engine, which helps developers build ML models for any kind of data. However, this does assume that you have a certain amount of ML expertise and experience, which is the sticking point.
After all, machine learning specialists are some of the most highly-sought after (and highly compensated!) developers out there. Not every company will have access to top AI talent or the budget to develop their own machine learning models. That’s where Cloud AutoML comes into play.
Cloud AutoML is fairly similar in its goals to the rest of Google’s ML offerings: bringing AI to developers and enterprises alike. It differs, however, by lowering the barrier to entry to developers with limited ML expertise. Now, companies with limited ML experience can utilize high-quality models that come with easily trainable models for fun and profit.
The first product released is AutoML Vision. It’s a simple GUI that allows users to train, evaluate, improve, and deploy models for image recognition. Google boasts that you can start training a model with as few as a dozen or so photographic samples, which is something indeed. AutoML Vision is able to do this thanks to shared use of popular public datasets like CIFAR and Image Net.
It’s apparently quite easy to generate high quality data thanks to the help of image labels (or the use of Google’s human labelling service). Turnaround is quite fast as well, with a simple model taking only minutes to pilot or a production-ready model in a day.
Auto CloudML is not exactly rocket science: the whole process is done through a drag and drop interface. Google is going to be doing all the work to train the model for the user, with no need for a ML specialist.
Of course, Cloud AutoML is fully integrated into other Google Cloud services, including Cloud Storage. While AutoML Vision is the first product out on the market, Google promises that there will be services for all the other major fields of AI coming soon. So, we should expect to see offerings for speech, translation, video, and more sometime soon.
If you’re interested in trying out Cloud AutoML, you can apply to try it free here. There’s no initial charge. But, fair warning, if you want to keep using it after your trial ends, you will have to pay for the privilege and a subscription to Google Cloud Platform ain’t cheap.