Over the years, the Google Brain team has focused on intelligent machine learning to improve people's lives. To this end, both in the field of pure theoretical research, or the research results into practical application, exploration never stopped. Recently, Google Brain team leader Jeff Dean in 2016 in various fields of the significant progress has been summarized.
Theoretical research results: Published 27 papers
In 2016, Google Brain team participated in the ICML, NIPS and ICLR and other international top machine learning conference, and in these meetings published 27 papers. In addition to professional meetings in the field of machine learning, many papers have been accepted by conferences in other fields, such as the ACL and CoNNL conferences in the field of natural language processing and the ICASSP conference in the speech field.
Google Brain also presented 34 papers to the ICLR 2017 for further study and discussion at the forthcoming conference.
Natural Language Understanding: 85% Better Translation Accuracy
Let the computer better understand the human language, Google Brain team has been a research focus. As early as 2014, three researchers at Google Brain demonstrated the potential of neural network technology to be used in machine translation through a related paper. The 2015 study also showed that this technique could be applied to image translation, sentence component analysis and Solve the problem of aggregation operations and so on.
In 2016, an end-to-end learning system was applied to Google Translate, which resulted in a significant increase in translation accuracy, with some language pairs achieving an 85% increase in accuracy, in close collaboration with the Google translation team. With the release of the multi-language translation system, so that the translation of the language to achieve more and more, to eliminate the language barrier goal a step further.
Three Ways to Improve Robot Autonomous Learning Efficiency
It is difficult to embed the new functions and new capabilities into the original robot, which is a difficult point for the traditional machine technology which is carefully designed by the algorithm and programmed by the human, and the machine learning technology provides the breakthrough point for this difficulty. In 2016, Google Brain and Google X team, through a large number of experiments to explore the robot arm in a shorter period of time to self-study and training "hand-eye coordination" approach.
The Google Brain team finally came up with three ways to make robots get new capabilities: reinforcement learning, interactive learning, and demonstration learning. The team will continue to study the robot in a more complex reality of the environment to acquire new skills.
To assist in the diagnosis of disease
The great potential of the application of machine learning technology in the medical field is very exciting. The Google Brain team's paper in the Journal of the American Medical Association (JAMA) demonstrates the potential of machine learning techniques to be used in the diagnosis of diabetic retinopathy, which will benefit patients in areas that are not diagnosed and treated.
Google Brain believes that machine learning, both from the quality or efficiency level, can greatly enhance the patient's treatment experience. In 2017, research in this area will be further strengthened.
Machine learning and the violent collision of art
The Magenta Project, launched in 2016, explores how machine learning systems can be used to learn about human creativity and seek the spark of art and machine intelligence collisions.
Magenta is becoming the most advanced content creation generation model, starting with the generation of music and images, into text and VR generation.
Ensures the safety and fairness of artificial intelligence
Complex and powerful artificial intelligence system will be more and more involved in people's lives, its security and fairness must be strictly guaranteed. Google Brain and Stanford University, Berkeley and OpenAI and other universities and institutions to jointly publish papers, discusses the field of artificial intelligence security should be the focus of attention. At present, it is important to ensure the confidentiality of machine training data.
Google Brain is also working to make artificial intelligence systems for complex decisions. In this process, to ensure the fairness of decision-making is also very important. In a paper that oversees the fairness of the machine learning process, Google Brain discusses how technology can help prevent bias and discrimination.
Open source project TensorFlow is popular
TensorFlow, the most popular machine learning project on GitHub, attracted more than 570 contributors for more than 10,000 offers. At the same time, TensorFlow has also been adopted by many well-known research teams and enterprises, and put into practical application: from looking for rare manatees, to help farmers choose cucumber, various projects have its presence.
In November 2016, TensorFlow open source one year, a paper on the TensorFlow computer system in the computer system research conference OSDI was submitted to explain its large-scale machine learning principle.
Building a machine learning community, and promoting education and technological development
In order to accelerate the development of the machine learning field, cultivate more professionals, Google Brain has also been actively building a machine learning community to carry out machine learning research.
Google Brain promotes education and technology development in the field of machine learning by providing free depth learning web-based courses, creating visual learning systems, online quizzes, and internships.
Let machines learn all over Google
Inside Google, the wave of machine learning is also sweeping every corner. Machine learning expertise and awareness have been consciously infiltrated into many teams and applied to product development. For example, the Google Brain team has significantly improved machine learning efficiency and performance for Google's custom accelerator ASICs and has been applied to the neural network machine translation system and AlphaGo and Li Shi-shi's Go game.
All in all, 2016 is a very exciting year for the Google Brain team, colleagues and partners. It is expected that the study of machine learning in 2017 will have a far-reaching impact on the world.