What machine learning is, how it already makes our lives easier, and where it will take us.
(+ fun experiments you can do from your PC )
Not long ago it might have seemed like science fiction, now at least future. But machine learning, is already, increasingly, present in our lives. And it is not even a new science, because the concept of machine learning has its roots in the 18th century, in the field of statistics.
So machine learning means machines - computers - capable of learning without being explicitly programmed to do so .
The operation of machine learning
From a series of examples, the machine captures the patterns (i.e., the patterns)characterizing them and then uses them to make predictions about other similar examples.
Let's take the word "dog" in analysis: once it has collected N examples of photos labeled "dog" (and so for millions of other labels), the computer goes in search of the patterns of pixels and colors that will help it figure out whether there is a dog in the photo; it tries to guess which patterns are right for identifying dogs and continues its study by looking for an example of a picture of a dog to see if the patterns work.
These operations are repeated about a billion times: the machine looks at an example and, if the result is incorrect, changes the patterns it is using to get a better result. In the end, the patterns form a machine learning-based pattern, a deep neural network, that can (almost always) correctly identify dogs, cats, children, places, sunsets, and many other things.
Machine learning is already present in most of Big G's products, including Search, Gmail, YouTube, Maps and Android.
Within Google Photos, for example, you can search for anything, "hug" or "dog," and you will find all the photos that contain them. The system has been trained to recognize images within photos and videos.
Thanks to Google Translate, it is possible to speak, write or read in more than 100 languages. Google is pioneering the use of machine translation, which uses statistical models to translate texts. In the past 6 months, Translate has further improved with machine translation based on neural networks. The neural system translates whole sentences instead of translating piece by piece: it uses a broader context to help itself choose the most relevant translation and refines it to achieve a result more like human speech.
With theGoogle app, all you have to do is say "Ok Google" on your smartphone to start searching and find the answers you are looking for. Speech recognition turns sounds into words, natural language processing aids understanding of meaning, and Search ranking shows links with the best results.
You can respond to emails by simply touching a button: the SmartReply feature of Inbox suggests a response so that you can respond to emails with a gesture. Gmail and Chrome also protect you by filtering spam and malware.
---
For the foreseeable future, the whole industry is working to make machine learning faster, with fewer examples.
One method-and this is what Google is particularly focusing on-is 'regularizing' the machines, that is, equipping them with more 'common sense,' so that when faced with an example that is slightly different from others-a dog wearing a cowboy hat is still a dog-the machine is not misled.
---
Google Research Europe, a European research team dedicated to Machine Learning
Google's ongoing research in the field of Machine Learning underpins many of the products used by hundreds of millions of people every day. Big G's goal is to create products that can improve people's daily lives and, why not, help tackle the biggest problems. To do this, it uses a variety of computer science techniques, one of the most important of which is definitely the machine learning.
What makes continued progress in the field possible is the enormous collaboration among Google researchers around the world, who bring together a unique wealth of information and ideas on cutting-edge technologies and methods in the field of Machine Learning, with the goal of developing useful tools and products.
This is why Google Research Europe was born in June 2016, based in the company's Zurich offices.
Zurich is also home to Google's largest research center outside the United States. It is here that the engine behind Knowledge Graph and the conversation engine behind the Google Assistant in Allo were developed.
In addition to pursuing collaboration with Google's various research teams, the European research team focuses on three core areas: machine learning; natural language learning and processing; and artificial perception. In these areas, the team actively works on improving the machine learning infrastructure, thereby facilitating research and subsequent practical application. In addition, researchers in the Zurich office work closely with the team of linguists, with the goal of advancing natural language learning in collaboration with Google's research groups around the world.
Through machine learning Google is also helping to address some important issues. Here are some recently announced projects:
- Helping pathologists diagnose cancer through deep learning
Breast cancer diagnosis and subsequent treatment decisions can be highly variable, which is not surprising given the enormous amount of information that must be evaluated in order to make an accurate diagnosis.
Google is how deep learning can be applied to support pathologists by creating an automatic identification algorithm that can complement their work. - Saving energy in data centers
To reduce energy consumption at Google. Machine learning is being used: thanks to neural networks, data center operations are being optimized and their consumption reduced like never before. - Identifying Gender Bias in Film
Oscar winner Geena Davis wondered whether unconscious biases toward women lead people to accept that female characters have fewer lines and less visibility than men in movies. So in 2007 she founded the Geena Davis Institute on Gender in Media and began collecting data. It was long and challenging work: a team of researchers started by watching one movie at a time and noting patterns based on gender. Then machine learning, the ideal tool for understanding and interpreting huge amounts of complex data, entered the picture. Today, software accurately measures how often women are seen or heard on screen. The tool learned to recognize different characters on screen, determine their gender, and calculate how often and for how long they speak to each other. The results were significant: men are seen and heard almost twice as often as women; women are underrepresented in all categories, including the 100 highest-grossing films of the past three years; women are barely seen in Oscar-winning films: they account for 32 percent of pictures and 27 percent of speech.
Google's machine learning tools are open.
The flexible machine learning system TensorFlow(www.tensorflow.org) is open sourcetoallow everyone to develop their own models. Google also offers several simplified solutions, such as Speech, Vision, Translate API and the TensorFlow service managed through the Google Cloud machine learning platform(https://cloud.google.com/ml).
For more information: http://research.google.com/pubs/MachineIntelligence.html
And now... enjoy Google's A.I. Experiments, which show how the machine can learn to understand many of the things around us.
In Quick, Draw!, a neural network tries to recognize what you are drawing, all using the same technology by which Google Translate recognizes handwriting. And the more you play, the more the system learns.
With A.I. Duet, however, you can improvise a piano duet with the computer. Try playing a few notes, and the computer will respond with a melody.
Or try Giorgio Cam: take a picture and the computer will tell you what it sees by turning it into the words of a song, perhaps in rhyme, to the tune of a song by the great Giorgio Moroder.
As recently done in this post, we remind you that FRIDAY, OCTOBER 19, GOOGLE WILL BE THE OPENING GUEST OF THE MASHABLE SOCIAL MEDIA DAY ITALIA + DIGITAL INNOVATION DAYS.
Marianna Ghirlanda - Head of Creative Agencies at Google Italia - will explain to the audience How to be successful thanks to YouTube, without magic formulas, but with a lot of creativity and the ability to surprise with unexpected solutions.