Google is known for experimenting heavily with new technologies, but it’s becoming clear that machine learning is becoming a real value proposition for the search giant.

On Thursday, at the Google Cloud Platform Next conference, Google’s Julia Ferraioli broke down some of the key announcements that Google made around machine learning the day prior, and how developers could take advantage of them.

Machine learning has struggled with a formal definition, but Ferraioli said that Google views it as how developers can add intelligence to their applications. And, using that data and intelligence is growing in importance.

“How well you use your data can determine the degree of your success,” Ferraioli said.

SEE: Machine automation policy guidelines template (Tech Pro Research)

Machine learning has long been the realm of a few specialized individuals and used to require education at the PhD level to implement. At its core, machine learning is built on three components: Data, algorithm, and insight.

An example of machine learning in action can be seen in Google Photos, where Google recently added the capability to search within the photos themselves with no manual labeling necessary. Google Translate also uses machine learning.

So, how does Google approach machine learning overall? Ferraioli said it’s a spectrum.

TensorFlow is an open source machine learning library. Released last year, it operates on tensors, or dimensional arrays, and it uses a flow graph to explain the data. TensorFlow is on the academic or research side of machine learning at Google.

SEE: Managing AI and ML in the enterprise (ZDNet special report) | Download the report as a PDF (TechRepublic)

Machine learning APIs are on the opposite side of that spectrum and require much less understanding of machine learning to implement within an application. Cloud Machine Learning, announced Wednesday, is in the middle and can extend to either side.

Ferraioli said developers can use Cloud Machine learning “When you have a customized problem that you want to solve.”

Cloud Machine Learning is a fully managed service, and developers can train it using a custom TensorFlow graph. It offers batch and online prediction at scale and an integrated Datalab experience, but regression and classification are its two primary tasks.

The example that Ferraioli used was a program called “Can I Hug That?” The application uses images and labels, as well as a trained classifier to answer the question. JSON test and train files help the machine learning tool better identify and classify the images, and a hyperparam file shows how many labels you have and other aspects of the network.

On the API side of things, Google is well known for its Google Translate API, but it also has a new Speech API that it released Wednesday and its Cloud Vision API.

The new Speech API allows users to pass raw audio data and language information, and it will return a transcript of the audio data. It works with roughly 80 languages and offers a streaming or non-streaming response.

The Cloud Vision API can detect faces, landmarks, logos, and text, among other images. It can also perform sentiment analysis and is a straightforward REST API. The API works on a base64-encoded image, connects to Google Cloud Storage, and will return a label and a score pair.

SEE: Google Cloud Platform signs up enterprise giants, how does it compare to AWS? (TechRepublic)

David Zuckerman from Wix joined Ferraioli on stage to explain how his company, which helps people build websites using a freemium model, has used the Cloud Vision API.

Zuckerman’s problem was that he, and his company, wanted to know more about their users, or help drive them to convert to a paid site, but they couldn’t directly ask them because they found that it would turn them off to the service. As they approached the problem, machine learning came up as a possible solution.

“If we can drive a car autonomously, we can definitely learn more about our users with machine learning,” Zuckerman said.

However, machine learning is traditionally a large investment, and you need the right people to be able to properly train the models. Wix didn’t have the right people and couldn’t afford that kind of investment.

Working with Google, Wix got early access to the Cloud Vision API and now they can search and label all user images that are uploaded, as well as perform face detection on all photos. They were essentially able to use machine learning without knowing anything about it, Zuckerman said.

As machine learning and the concept of adding intelligence to all business processes continues to increase, Google’s investment in machine learning could pay off big in the enterprise.

The 3 big takeaways for TechRepublic readers

  1. Google’s approach to machine learning is a three-tiered spectrum with its TensorFlow tool, the new Cloud Machine Learning tool, and its machine learning APIs.
  2. On Wednesday, Google announced the new Cloud Machine Learning tool, as well as the new speech API, which transcribes audio in almost 80 different languages.
  3. As corporations and developers continue to strive to build intelligent applications and processes, Google’s investment in machine learning could return dividends in the enterprise.