Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation

Publication Year

2016

Authors

Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V. Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, Jeff Klingner, Apurva Shah, Melvin Johnson, Xiaobing Liu, Łukasz Kaiser, Stephan Gouws, Yoshikiyo Kato, Taku Kudo, Hideto Kazawa, Keith Stevens, George Kurian, Nishant Patil, Wei Wang, Cliff Young, Jason Smith, Jason Riesa, Alex Rudnick, Oriol Vinyals, Greg Corrado, Macduff Hughes, Jeffrey Dean

Abstract

Neural Machine Translation (NMT) is an end-to-end learning approach for automated translation, with the potential to overcome many of the weaknesses of conventional phrase-based translation systems. Unfortunately, NMT systems are known to be computationally expensive both in training and in translation inference. Also, most NMT systems have difficulty with rare words. These issues have hindered NMT's use in practical deployments and services, where both accuracy and speed are essential. In this work, we present GNMT, Google's Neural Machine Translation system, which attempts to address many of these issues. Our model consists of a deep LSTM network with 8 encoder and 8 decoder layers using attention and residual connections. To improve parallelism and therefore decrease training time, our attention mechanism connects the bottom layer of the decoder to the top layer of the encoder. To accelerate the final translation speed, we employ low-precision arithmetic during inference computations. To improve handling of rare words, we divide words into a limited set of common sub-word units ("wordpieces") for both input and output. This method provides a good balance between the flexibility of "character"-delimited models and the efficiency of "word"-delimited models, naturally handles translation of rare words, and ultimately improves the overall accuracy of the system. Our beam search technique employs a length-normalization procedure and uses a coverage penalty, which encourages generation of an output sentence that is most likely to cover all the words in the source sentence. On the WMT'14 English-to-French and English-to-German benchmarks, GNMT achieves competitive results to state-of-the-art. Using a human side-by-side evaluation on a set of isolated simple sentences, it reduces translation errors by an average of 60% compared to Google's phrase-based production system.

Viewers can visit source article : https://research.google.com/pubs/pub45610.html

Microsoft and partners launch Intelligent Network for Eyecare in India

By Eden Estopace | 2017-01-04

inShare9

 Email PrintShare

Microsoft India and the L V Prasad Eye Institute recently launched the Microsoft Intelligent Network for Eyecare (MINE), a global consortium of commercial, research and academic institutions who have joined hands to apply artificial intelligence to deliver eye care services and help eradicate preventive blindness.

Partner organizations include Bascom Palmer – University of Miami, Flaum Eye Institute – University of Rochester (USA), Federal University of Sao Paulo (Brazil), and Brien Holden Vision Institute (Australia). 

These institutions vowed to  work on diverse datasets of patients across geographies to come up with machine learning predictive models for vision impairment and eye disease. This will include the rate of change of myopia in children, conditions that impact children’s eyesight, predictive outcomes of refractive surgery, optimal surgery parameters as well as ways to personalize a surgery and maximize its probability of success. 

By studying this data and applying advanced analytics with Microsoft machine learning technology to derive insights, MINE will aim to drive strategies to prevent avoidable blindness and help increase efficiency in the delivery of eyecare worldwide. 

Under the agreement, Microsoft will deploy its leading cloud platform technology – Cortana Intelligence Suite, for advanced analytics and to build Artificial Intelligence models on eyecare.

“By using our capabilities and pioneering work in the field of machine learning over the years, Microsoft has helped accelerate digital transformation in various key sectors in India, including agriculture and education,” said Anil Bhansali, Managing Director, Microsoft India (R&D) Pvt. Ltd., in a news release.

Dr. G.N. Rao, Founder-Chair, L V Prasad Eye Institute, said he is confident that partnership will not only open doors to opportunities in the field of eyecare, but also pave way for others to leverage technology to address several other critical eye diseases.

The partners said MINE will bring together different partner organizations on Azure and collectively work on ideas of impact. 

Viewers can read this article from source link : http://www.enterpriseinnovation.net/article/microsoft-and-partners-launch-intelligent-network-eyecare-india-463756281

Journalists watch a big screen showing live footage of the third game of the Google DeepMind Challenge Match between Lee Se-Dol and AlphaGo at a hotel in Seoul on March 12, 2016.  JUNG YEON-JE AFP/Getty Images

AlphaGo

Google’s AlphaGo AI Runs the Table on Asia’s Go Champs

David Z. Morris

Updated: Jan 07, 2017 1:18 PM Pacific

In a series of unofficial online games, an updated version of Google’s AlphaGo artificial intelligence has compiled a 60-0 record against some of the game’s premier players. Among the defeated, according to the Wall Street Journal, were China’s Ke Jie, reigning world Go champion.

The run follows AlphaGo’s defeat of South Korea’s Lee Se-dol in March of 2016, in a more official setting and using a previous version of the program.

Get Data Sheet, Fortune’s technology newsletter.

The games were played by the computer through online accounts dubbed Magister and Master—names that proved prophetic. As described by the Journal, the AI’s strategies were unconventional and unpredictable, including moves that only revealed their full implications many turns later. That pushed its human opponents into deep reflections that mirror the broader questions posed by computer intelligence.

“AlphaGo has completely subverted the control and judgment of us Go players,” wrote Gu Li, a grandmaster defeated by the program, in an online post. “When you find your previous awareness, cognition and choices are all wrong, will you keep going along the wrong path or reject yourself?”

Another Go player, Ali Jabarin, described running into Ke Jie after he had been defeated by the program. According to Jabarin, Jie was “a bit shocked . . . just repeating ‘it’s too strong’.”

For more on AI, watch our video.

Why Alphabet Shouldn't Have Ended its Robotics Unit

After purchasing it a few years ago

Predictably, the official statement from developers DeepMind is less angsty, emphasizing the potential for AI to “explore the profound mysteries of the game further in this spirit of mutual enlightenment”—and promising that AlphaGo will take its next-level game to more official venues soon.