Google announced two services today, one new and one out of preview. They are part of the company's ongoing push to fashion itself as a provider of not only tools for building machine learning resources, but also APIs for accessing premade ones.
Cloud Machine Learning (CML) can plug into Google's other storage, querying, and data-handling products to generate machine learning models. Among the data sources is Google Cloud Dataproc, the managed Hadoop and Spark platform that was previously announced but is now in general availability.
You may have been wondering when machine learning as a service would arrive in Google Cloud, considering it has been available on Amazon for months and on Azure for a year. CML is based on Google's open-sourced TensorFlow framework. TensorFlow, Google claims, was used to build and deliver many existing Google products with machine intelligence aspects, such as its speech-recognition API, newly available to the public.
Models built with TensorFlow outside of Google's services can be used with CML, but Google wants prospective users to do the data ingestion, management, and training directly on its cloud whenever possible.
Also among Google's offerings is a number of pretrained APIs for translation, machine vision, and speech recognition. IBM has similar items via Watson on Bluemix, and while Google's range is smaller, it features more products that are likely to find immediate uptake and use.
Google's been in a prime position to be a machine learning master, not only because it has generously shared open source machine learning creations like TensorFlow or built services atop existing, well-understood projects like Spark. Rather, it's due to Google's enviable position as the hub through which flows a staggering amount of the world's data -- raw material that can be used, in some form, to train Google's systems.
What Google is offering now is an option for people to bring their own data to the tooling it developed to make sense of its data trove. Apart from needing a source of training data, machine learning has also been pegged as difficult to use. The new wave of open source ML projects has eased the burden somewhat -- one of Spark's big appeals is how straightforward it is for developers.
However, Google aims to lower the hurdles by removing the hassles around building the infrastructure for such projects, and by providing an environment that's as elegant and as no-nonsense as the rest of its cloud is shaping up to be.