Google Cloud AutoML: no longer just bots-for-boffins

COMMENTARY
 
This week saw Google announce early alpha access to its Cloud AutoML service, designed to bring custom machine learning models to the general developer population (i.e. those without any specific deep learning expertise – not just deep learning researchers, or the data scientist community).

By Google’s reckoning, that opens up the power of deep learning to twenty times the number of people currently qualified to exploit the technology. Yes, there have been standardised building blocks for tightly-bounded use cases available in the market for a while now, but if you wanted to tailor the model beyond basic configurations you’d quickly find yourself needing the services of a deep learning expert.

So, like AWS a couple of months earlier (with its launch of Amazon SageMaker, which I covered in my report of re:Invent 2017), Google is now also on a mission to democratise access to machine learning.

Cloud AutoML is a suite of products designed to help developers with limited ML expertise leverage Google’s Neural Architecture Search and Transfer Learning technologies to train, evaluate, improve, and deploy their own custom models using a simple graphical user interface – customers bring their own training data; AutoML then creates the ML model. It’s ML-driven tools for automated ML model creation, if you like.

First out of the gate is AutoML Vision, for training custom vision models; Google promises more, “for all other major fields of AI” soon.

Why this, why now? With machine learning fast becoming a universally-offered component of any digital service (whether back-office or customer-facing), demand for qualified deep learning experts to help develop and deploy custom smart applications and online services has been outpacing supply. Only those with the deepest of pockets have been able to do anything other than integrate off-the-shelf components in common use cases. What AWS (with SageMaker) and Google (with AutoML) are seeking to do, is open up deep learning differentiation to the mass market.

So which should you choose? That largely depends on where the centre of gravity lies for the rest of your cloud service investments (and where your data is). Each vendor’s service is well integrated into the rest of its own cloud offerings, meaning that if you’re already an AWS shop, you’ll likely plump for Sagemaker and the AWS suite of ML services; if you’re more in with Google however, then it’s the AutoML tools that will provide you an easier route to ML deployment.

If you’re yet to commit to a primary cloud for your data and compute however, Google’s announcement marks an interesting play. As I commented in November 2017, the company’s getting much more serious about setting out Google Cloud Platform’s enterprise-friendly credentials (after years of being seen more as the super-smart, but not necessarily super business-focused, cloud option for high-end data-wrangling). It’s keen now to promote a digital transformation powered by Google Cloud as one that leverages its cutting-edge deep learning and machine learning expertise to transform what a customer can do with its business.

Of course, AutoML Vision only covers one aspect of ML, and it’s still only in Alpha, so this won’t smarten up your service stack overnight… but it is an important step in the democratising of access to the sort of deep learning techniques that have been giving Google’s businesses their edge for years. The bar just moved, and soon a lot more people will be able to reach it.

COMMENTARY

Leave a Reply

Your email address will not be published. Required fields are marked *