Twitter
Advertisement

Google has just open sourced the artificial intelligence engine that drives its core services

Machine learning development has just been put on a superfast track

Latest News
article-main
FacebookTwitterWhatsappLinkedin

Google, at its core, is a company driven by analytics. From the Web searches we throw at it to the manner in which we use Gmail, to plotting routes on Google Maps, the company has built an empire on analyzing massive amounts of data. In recent years, the company has made huge strides in not just analyzing this data but in intelligently predicting what users might need (Google Now,) and even assist in areas like breaking language barriers (Google Translate.) For anyone who’s used these services, you’ll notice they are getting increasingly adept at delivering surprisingly accurate and timely information, often belying the fact that no human powers these systems: all of this is based on artificial intelligence (AI.)

Virtually all of the large software and social media players are striving towards integrating some form of AI into their products and service, including Facebook, Twitter and Microsoft. In fact, many of these websites have either assimilated AI prowess from third parties, or even absorbed the talent behind the creation of this technology. For example, Torch: a system that was developed by researchers at New York University, plays a big role in driving Facebook’s News Feed, and several of its creators are currently employed with the company.

The specific area of AI that Google uses to power these groundbreaking services is known as Deep Learning. It is the system that makes it possible for Google’s apps and services to improve its data processing ability based on experience from past learnings. And it is used across media and applications; to recognize spoken words across dialects and languages, recognizing photos and faces to find other matches, and delivering more relevant Web search results.

The name of Google’s Deep Learning system is TensorFlow. And today, Google just took the step of open sourcing it. Yes, the actual code that powers all of their intelligence is now available for anyone to study, use and develop, free of charge. The implications of this move are massive for the progress and adoption of AI.

What is Deep Learning?

Deep learning is based on a field of computer science called Neural Networks, which strives to artificially mimic the functioning of the human brain. In traditional computing systems, tasks are often repetitive where the computer largely performs computations repeatedly and very quickly. With Neural Networks however, the system’s software is designed such that it can actually ‘learn’ from its own iterations, while autonomously generating more efficient approaches in solving a given problem--much like the human brain does when learning a new task or faced with a challenge.

In the training stage--much like when a child is learning to walk or when a person learns a new skill--these Deep Learning systems are fed with massive amounts of data that it analyses, learns from, and uses to make subsequent decisions and choices in a recursive learning process. Once accuracy reaches a high enough level, these learnings can be distilled into other systems that don’t require the entire power of the Neural Network. A good example of this is the Google Translate app, which can actually run standalone on a smartphone or tablet: Google’s Deep Learning systems do the heavy lifting of recognizing languages, then funnel these learnings into the app that even works offline.

Given the massively parallel nature of the data these Deep Learning systems work with (hundreds of streams of information that need to be processed simultaneously and in relation to each other,) computer scientists have discovered that Graphics Processing Units (GPUs, typically used in game consoles and visual rendering systems) have proved to be significantly more effective at Machine Learning compared to traditional CPUs. Google, for example, uses massive arrays of GPUs in their Deep Learning systems.

“What we’re hoping is that the community adopts this as a good way of expressing machine learning algorithms of lots of different types, and also contributes to building and improving [TensorFlow] in lots of different and interesting ways,” said Jeff Dean, one of Google’s key engineers who’s been driving their deep learning initiatives.

TensorFlow is built using the popular C++ programming language, so coders can immediately use both C++ or Python, a popular language that is used to build a range of desktop to mobile apps along with Web services. With the system being open sourced, TensorFlow is now likely to be extended to other programming languages including Google Go, Java and even Javascript.

Google will be managing this new open source project via the website Tensorflow.org, which is replete with tutorials and documentation. The Tensorflow code is available under the Apache 2 license, which grants the freedom to utilize the code as a person pleases.

To be clear, Google isn’t opening every part of its AI engine to the world--only some algorithms are being open sourced, which in itself is significant given that they haven’t done so in the past. Also the algorithms alone are one part of the AI engine; the other large part being the massive hardware infrastructure that facilitates this AI horsepower still remains proprietary.

Still, with one of the world’s most advanced Machine Learning systems now available to the general public we can expect to see huge strides in the space, not just within Google’s own portfolio of products but in other applications and services as well.

Find your daily dose of news & explainers in your WhatsApp. Stay updated, Stay informed-  Follow DNA on WhatsApp.
Advertisement

Live tv

Advertisement
Advertisement