Google Brain is a deep learning artificial intelligence research project at Google. It combines open-ended machine learning research with system engineering and Google-scale computing resources.
Video Google Brain
Mission
Google Brain's mission is to improve people's lives by making machines smarter. To do this, the team focuses on constructing models with high degrees of flexibility that are capable of learning their own features, and use data and computation efficiently.
As the Google Brain Team describes "This approach fits into the broader Deep Learning subfield of ML and ensures our work will ultimately make a difference for problems of practical importance. Furthermore, our expertise in systems complements this approach by allowing us to build tools to accelerate ML research and unlock its practical value for the world."
Maps Google Brain
History
The so-called "Google Brain" project began in 2011 as a part-time research collaboration between Google Fellow Jeff Dean, Google Researcher Greg Corrado, and Stanford University professor Andrew Ng. Ng had been interested in using deep learning techniques to crack the problem of artificial intelligence since 2006, and in 2011 began collaborating with Dean and Corrado to build a large-scale deep learning software system, DistBelief, on top of Google's cloud computing infrastructure. Google Brain started as a Google X project and became so successful that it was graduated back to Google: Astro Teller has said that Google Brain paid for the entire cost of Google X.
In June 2012, the New York Times reported that a cluster of 16,000 computers dedicated to mimicking some aspects of human brain activity had successfully trained itself to recognize a cat based on 10 million digital images taken from YouTube videos. The story was also covered by National Public Radio and SmartPlanet.
In March 2013, Google hired Geoffrey Hinton, a leading researcher in the deep learning field, and acquired the company DNNResearch Inc. headed by Hinton. Hinton said that he would be dividing his future time between his university research and his work at Google.
On 26 January 2014, multiple news outlets stated that Google had purchased DeepMind Technologies for an undisclosed amount. Analysts later announced that the company was purchased for £400 million ($650M USD / EUR486M), although later reports estimated the acquisition was valued at over £500 million. The acquisition reportedly took place after Facebook ended negotiations with DeepMind Technologies in 2013, which resulted in no agreement or purchase of the company.
Recent breakthroughs
Artificial-intelligence-devised encryption system
In October 2016, the Google Brain ran an experiment concerning the encrypting of communications. In it, two sets of AI's devised their own cryptographic algorithms to protect their communications from another AI, which at the same time aimed at evolving its own system to crack the AI-generated encryption. The study proved to be successful, with the two initial AIs being able to learn and further develop their communications from scratch.
In this experiment, three AIs were created: Alice, Bob and Eve. The goal of the experiment was for Alice to send a message to Bob, which would decrypt it, while in the meantime Eve would try to intercept the message. In it, the AIs were not given specific instructions on how to encrypt their messages, they were solely given a loss function. The consequence was that during the experiment, if communications between Alice and Bob were not successful, with Bob misinterpreting Alice's message or Eve intercepting the communications, the following rounds would show an evolution in the cryptography so that Alice and Bob could communicate safely. Indeed, this study allowed for concluding that it is possible for AIs to devise their own encryption system without having any cryptographic algorithms prescribed beforehand, which would reveal a breakthrough for message encryption in the future.
Image enhancement
In February 2017, Google Brain announced an image enhancement system using neural networks to fill in details in very low resolution pictures. The examples provided would transform pictures with an 8x8 resolution into 32x32 ones.
The software utilizes two different neural networks to generate the images. The first, called a "conditioning network," maps the pixels of the low-resolution picture to a similar high-resolution one, lowering the resolution of the latter to 8×8 and trying to make a match. The second is a "prior network", which analyzes the pixelated image and tries to add details based on a large number of high resolution pictures. Then, upon upscaling of the original 8×8 picture, the system adds pixels based on its knowledge of what the picture should be. Lastly, the outputs from the two networks are combined to create the final image.
This represents a breakthrough in the enhancement of low resolution pictures. Despite the fact that the added details are not part of the real image, but only best guesses, the technology has shown impressive results when facing real-world testing. Upon being shown the enhanced picture and the real one, humans were fooled 10% of the time in case of celebrity faces, and 28% in case of bedroom pictures. This compares to previous disappointing results from normal bicubic scaling, which did not fool any human.
Google Translate
The Google Brain Team has recently reached significant breakthroughs for Google Translate, which is part of the Google Brain Project. In September 2016, the team launched the new system, Google Neural Machine Translation (GNMT), which is an end-to-end learning framework, able to learn from a large amount of examples. While its introduction has greatly increased the quality of Google Translate's translations for the pilot languages, it was very difficult to create such improvements for all of its 103 languages. Addressing this problem, the Google Brain Team was able to develop a Multilingual GNMT system, which extended the previous one by enabling translations between multiple languages. Furthermore, it allows for Zero-Shot Translations, which are translations between two languages that the system has never explicitly seen before. Recently, Google announced that Google Translate can now also translate without transcribing, using neural networks. This means that it is possible to translate speech in one language directly into text in another language, without first transcribing it to text. According to the Researchers at Google Brain, this intermediate step can be avoided using neural networks. In order for the system to learn this, they exposed it to many hours of Spanish audio together with the corresponding English text. The different layers of neural networks, replicating the human brain, were able to link the corresponding parts and subsequently manipulate the audio waveform until it was transformed to English text.
Robotics
Different from the traditional robotics, robotics searched by the Google Brain Team could automatically learn to acquire new skills by machine learning. In 2016, the Google Brain Team collaborated with researchers at X to demonstrate how robotics could use their experiences to teach themselves more efficiently. Robots made about 800,000 grasping attempts during research. Later in 2017, the team explored three approaches for learning new skills, i.e., through reinforcement learning, through their own interaction with objects, and through human demonstration. To build on the goal of the Google Brain Team, they would continue making robots that are able to learn new tasks through learning and practice, as well as deal with complex tasks.
In Google products
The project's technology is currently used in the Android Operating System's speech recognition system, photo search for Google+ and video recommendations in YouTube.
Team and location
Google Brain was initially established by Google Fellow Jeff Dean and visiting Stanford professor Andrew Ng (Ng later left to lead the artificial intelligence group at Baidu). In 2014, the team includes Jeff Dean, Geoffrey Shields, Greco Rado, Quoc Le, Ilya Sutskever, Alex Kelly Forth Alex Krizhevsky, Samy Bengio and Vincent Vanhoucke. At present, numbers of Chinese were in the position. In 2017, team members include Anelia Angelova, Samy Bengio, Greg Corrado, George Dahl (ML Researcher), Michael Isard, Anjuli Kannan, Hugo Larochelle, Quoc Le, Chris Olah, Vincent Vanhoucke, Vijay Vasudevan and Fernanda Viegas. Chris Lattner, who created Apple's new programming language Swift and then ran Tesla's autonomy team for six months joined Google Brain's team in August 2017.
Google Brain is based in Mountain View, California and has satellite groups in Cambridge, Massachusetts, London, Montreal, New York City, San Francisco, Toronto, and Zurich.
Career opportunities
Google Brain Residency Program
Google Brain Residency Program is targeted at people who are eager to devote own passion to machine learning and artificial intelligence. This is an opportunity to get hands-on experience in Google team and have chance to keep in touch with professional researchers with Google Brain team. The program lasted 12 months.
Within the program were groups of new graduates from top universities with degree of BAs or Ph.Ds in computer science, physics, mathematics, and neuroscience, or others who come from years of industry experience. They were picked to work with researchers in Google Brain Team at the forefront of machine learning.
The breadth of topics in this program allowed team members to flexibly combine their professional knowledge with the application of algorithms, natural language understanding, robotics, neuroscience, genetics and more. Just several months, these new future researchers have already made a great impact in the research field.
Some of the recently published technical papers resulting from the residency program are listed below:
Unrolled Generative Adversarial Networks
Conditional Image Synthesis with Auxiliary Classifier GANs
Regularizing Neural Networks by Penalizing Their Output Distribution
Mean Field Neural Networks
Learning to Remember
Towards Generating Higher Resolution Images with Generative Adversarial Networks
Multi-Task Convolutional Music Models
Audio DeepDream: Optimizing Raw Audio With Convolutional Networks
Reception
Google Brain has received in-depth coverage in Wired Magazine, the New York Times, Technology Review, National Public Radio, and Big Think.
See also
- Artificial intelligence
- Glossary of artificial intelligence
- Google X
- Quantum Artificial Intelligence Lab - run by Google in collaboration with NASA and Universities Space Research Association
- Google Translate
- TensorFlow
References
Source of article : Wikipedia
0 Comments