The Future Is Now: Machines Track Your Data to Learn How You Think

Everything you need to about Google’s Machine Learning technology, and how it works.

‘Let your machines learn from your data, and start solving problems for you’. That’s how my tête-à-tête with Mohit Pande, Google India’s Cloud head began. Google is at the forefront of solving millions of problems globally, with the help of ‘Machine Learning’ (ML).

Machine Learning gives computers the ability to learn without being explicitly programmed.

Just How Does ML Work?

The process refers to programming computers/machines that are coded to think like human beings – human decision-making processes are encoded into algorithms that are then used by these computers. The machines can then be connected to the internet, allowing them access to vast tracts of data from across the world.

ML devices use ‘neural networks’ to carry out their functions – these are computer systems, designed to function in much the same way a human brain does with a network of neurons. With the help of these ‘neural networks’, machines are capable of making decisions, detecting patterns, and predicting events, based on the data they are fed.

You show the machine 10,000 pictures of a cat, and by the time you show it the 10,001st picture, it will say it’s a cat. This way, the machine learns to make a decision and then classify it. The more the data, the better the output.

Mohit Pande, Country Head - Cloud & Business Innovation, Google India

Here’s a hypothetical situation that could play out. It’s one of those numerous Whatsapp forwards, but does show how ML can (one day, perhaps...) take over our lives.

What Can We Use It For?

When quizzed about the problem-solving that Google uses ML to enhance, Pande said that Google’s large pool of user data allows them to provide answers to text, voice, speech and translation-related issues posed by users.

But that’s not all. ML applications currently in the works can also read text and detect the tone of what is being written. For instance, they can figure out if a user is congratulating someone or complaining about something, and act accordingly.

Recently, Google has given us a real-world example of ML put to work, with products like Suggested Sharing, and Photo Books that utilise it to select the best of your photos and make a photo album for you.

Using our Machine Learning technology, Google Photos will not only remind you to share, it will automatically select the right photos, and even suggest who you should send them to based on who was in the photos.

Google blog post

Google Assistant is one of the byproducts of Machine Learning, that helps you with answers to even the most basic day-to-day problems. 

Mohit Pande, Country Head – Cloud & Business Innovation, Google India

Pretty soon, it’ll also play music to match your mood, or simply take commands from a person’s tone. This has become evident with devices like Google Home, which now understands multiple languages and answers appropriate, though that is only available in the West for now. Closer to home, you could use Google Assistant to sort through the photos on your smartphone and create customised albums.

Google’s large consumer footprint provides us with ample speed, voice, text and translation-related data, and we’re training that onto a machine model. 

Mohit Pande, Country Head – Cloud & Business Innovation, Google India

Google claims to deliver 70-80% accuracy while translating languages via speech-to-text at current.

Behind the Scenes

Machine Learning, very simply put, revolves around data, the algorithm, and the computing power of the machinery, says Pande.

To accelerate the pace of open ML research, Google has come out with its TensorFlow Research Cloud (TFRC). This is a cluster of 1,000 Cloud Tensor Processing Units (TPUs) that are made available free of cost, to support a broad range of computationally-intensive research projects that might not be possible otherwise, according to Google Research.

TFRC facilitates Machine Learning on systems that are capable of building and training neural networks.

Mohit Pande, Country Head – Cloud & Business Innovation, Google India

As shown through this chart, the human decision-making process is encoded into an algorithm, which is then fed into the TPUs to derive answers.

Google has open-sourced its ML Application Program Interfaces (APIs), so that the developer community can create applications and devise features to be used for problem-solving of all kinds. But the true test of its capability is to integrate the ML-based applications into our daily lives.

Google isn’t resting on its laurels here, which became evident when Sundar Pichai, CEO, Google announced the second generation Cloud TPUs at Google I/O 2017. This, Google claims, will take computing for ML to the next level, delivering quicker results than the systems are capable of right now.