Open the Google (microblogging) translation app, point the phone lens at a foreign language that you don't understand, and then automatically display the translation results on the screen.
Received an email from a friend. When checking the email, Gmail has prepared for you something that may need to be replied.
Whether it's Google Translate or Gmail, they have been launched for a long time and are not new products, but Google has and is continuing to "upgrade" where users can't see it. The above translations and the improvements provided by Gmail benefit from this.
“Upgrade†stems from the implementation and implementation of Google’s AI first strategy. In 2016, Google CEO Sundar Pichai announced that Google’s strategy has shifted from Mobile First to AI First, after years of investment in artificial intelligence technology. Now, Google is making its products smarter through technologies such as deep learning and machine learning.
Upgrading its products is just one of the goals that Google expects to achieve through AI. On November 28th, at Google's Asia Pacific Media Open Day in Tokyo, Japan, Google senior researcher and head of Google Brain, Jeff Dean, said that Google's vision in the field of artificial intelligence is to benefit everyone in three ways: Make Google products more useful, help businesses and developers innovate, and provide researchers with the tools to solve the major challenges facing humanity.
As can be seen from a series of products and cases displayed by Google, Google is working hard to achieve the above goals.
Google's new magicHow does AI make the product practical? The actual experience is the best way to answer this question.
At this year's Google Asia Pacific Media Open Day, Google has prepared a variety of products, including various image products, Google assistant, Google translate and a series of small applications that use AI technology such as machine learning.
Take the image application as an example. In real life, people often encounter scenes that need to convert paper-based images or files into an electronic version. Usually, users need to use professional scanning equipment to get the ideal electronic version, because if you use mobile phones When taking pictures, there will be problems such as reflection and distortion.
In response to this demand, Google launched an app called “Photo Scannerâ€. Users follow the guidelines to take five different angles of photos on the paper version. The app will calculate and synthesize the captured images, and finally output a Use a professional device to scan the same electronic version of the effect.
Photo scanners are just one example of Google's use of AI technology to improve the application experience. In addition to image applications, voice, text and other related products have different levels of experience improvement.
More than just software, Google is still trying to combine AI, software, and hardware with AI technology.
As an important hardware for Google's landing voice interactive products, Google Home can now offer a wide variety of services, such as its ability to recognize the voices of different users and give different feedback. For example, both A and B have a pet dog. When A tells Google Home that he wants to see a pet photo, Google Home can recognize A's voice and call A's pet photo; and when B asks for the same request. Google Home brings out a photo of B's ​​pet.
This feature is Google Home's Voice Match, which is enabled by the voice assistant to recognize different voices with the help of machine learning. According to reports, voice pairing currently supports up to six users connected to the same Google Home, Google Home is the first smart speaker on the market with this feature.
Machine learning has also been applied to the medical field. By working with doctors in India and the United States, Google created a dataset containing 128,000 fundus scans for training a deep neural network for detecting diabetic retinopathy.
After training, the professional accuracy of the model to identify fundus scans exceeds the average level of professional doctors, and this can help doctors improve the diagnostic efficiency and enable patients to get treatment as soon as possible.
At the same time, through TensorFlow, Cloud Machine Learning APIs and Tensor Processing Unit (TPU) computer chips, Google has opened AI capabilities to more developers. Food companies can use this to improve the efficiency of food inspections, and biologists can more effectively understand the habits of birds and improve the protection.
Whether it is to optimize existing products, open AI capabilities, and solve common problems for humans, the basic premise for achieving these goals is to consistently invest in AI research and make progress. In Google's AI strategy, machine learning is a top priority.
Change this happensMachine learning is a form of computer science. It is easier to write programs that enable computers to learn how to become intelligent, rather than writing intelligent programs directly. In general, the purpose of machine learning is to make the machine itself smart.
According to Jeff Dean, machine learning is Google's focus in the field of artificial intelligence. Google has been studying machine learning for a long time, but machine learning is still in its infancy. Today, machine learning is very helpful in four key aspects of classification, prediction, understanding, and generation. These features have been used in almost all of Google's products.
Whether it's Goolge Photos, Google Translate, Google Lens, Gmail, Inbox, Google Maps, or Google Assistant, YouTube, machine learning technology allows them to provide a better experience.
The new Pixel phone has a portrait mode, which can soften the background when shooting portraits. Under traditional technology, this requires a multi-lens professional camera, but the combination of machine learning and computational photography makes Pixel phones rely on The same effect is achieved by having one lens on each side.
Photographs without portrait mode (left) and photos using portrait mode (right)
According to Linne Ha, director of Google's search project, deep neural network technology greatly improves the accuracy of speech recognition in voice search, which allows users to talk freely with mobile phones in noisy environments. With the help of machine learning, natural language processing systems can better understand what you want to say. With the help of Project Unison, a project that uses machine learning to implement text-to-speech conversion, the phone can be used in languages ​​that are not rich in corpus, such as Bengali, Khmer and Javanese, through the conversion engine.
The actual results have proved that machine learning combined with specific applications can achieve very good results, but Jeff Dean still suggests two major challenges in machine learning today, first of all the accessibility of machine learning models; second, the inclusiveness of machine learning models. .
To solve the first problem, Google will offer free machine learning courses on the Internet next year. In order to solve the second problem, Google launched the People + AI Research (PAIR) program and established it in cooperation with the Geena Davis Institute. GD-IQ (a tool that uses machine learning to detect gender bias in movies).
The above measures have helped solve the challenges faced by machine learning, but for Google, under the strategic guidance of AI frist, the challenges are not limited to this.
From the perspective of industry competition, more and more companies have launched a machine learning open source platform. How does Google respond to competition and remain attractive to developers? From the perspective of competition among countries, Google is also facing the impact of the US government's willingness to invest in and support the development of the AI ​​industry.
Faced with industry competition, Jeff Dean said, “TensorFlow is constantly evolving and constantly adding new features. They may target different people, some for researchers, and some more for mobile platforms. So this competition is Ok. TensorFLOW's open source software has a very flexible Apache 2.0 licensing mechanism."
In the face of competition between countries, Google chose to establish a local team to improve the speed of AI development. Jeff Dean said that Google is forming an AI team in China, with teams mainly located in Beijing and Shanghai.
This approach is obviously from the perspective of new talents - "We want to focus on the next generation of people with better computing power, solve practical problems, interesting problems. We are not worried about competitors, we are concerned about our own research. ."
From research to application to openness, Google has formed a relatively complete puzzle under the guidance of AI first strategy. Although the development of artificial intelligence has caused some public concerns about its safety, Jeff has been involved in this industry. Dean seems that machine learning can help humans solve more complex problems. "We should now focus on solving immediate problems," and this will also affect Google's future development of AI.
a thing that joins together two parts of sth, two vehicles or two pieces of equipment.
Custom Coupling,Coupling Of Encoders,Useful Coupling,Latest Coupling
Yuheng Optics Co., Ltd.(Changchun) , https://www.yhencoder.com