This is not the martial arts world - she has done it. The 26-year-old Holgate got a second black belt for Taekwondo. This time is the black belt of the algorithm. Holgate spent weeks immersing himself in a program, and this time more powerful than melee training is machine learning. As an engineer in the Google Android division, Holgate is one of the 18 programmers of the Machine Learning Ninja Project this year. The members of the project are made up of talented programmers on the team. The members train according to the model in The Game of Ander. In order to teach them artificial intelligence technology to develop better products. Even this will make the software they write more difficult to understand.

ChrisTIne Robson is the product manager for Google's internal machine learning project. He manages the project. He said: "Our slogan is, do you want to be a machine learning ninja? We invite Google employees to join the machine learning team and spend 6 Study machine learning in the month, do some projects at the same time, and learn from the project practice, during which time the instructor will provide guidance."

Holgate came to Google four years ago, when she earned a degree in computer science and mathematics. This training opportunity is an opportunity for Holgate to master the most popular branch of the software field: with learning algorithms and large amounts of data, let The software completes the related tasks. For many years, machine learning has been seen as a discipline that belongs to only a few elites. But now that the era is over, machine learning is driven by neural networks, a way of mimicking how biological brains work. The latest research shows that machine learning can empower computers to harness human power and even beyond human power in certain areas. Google is committed to expanding its elite team internally and expects to turn it into a normal state. For engineers like Holgate, the Ninja project is an opportunity to be at the forefront of technology where they can learn the most advanced technology from the best engineers.

Holgate said: "These people are building a ridiculous model, and they all have a Ph.D. At first, I was very scared, but I learned to accept it."

Nearly half of Google’s 60,000 employees are engineers, so this is a small project. But the project symbolizes Google's cognitive transformation. Although machine learning has long been part of Google technology, and Google has hired many top experts in the field, Google seems to be more interested in this technology this year. At a conference call at the end of last year, Google CEO Sundar Pichai proposed a new corporate philosophy: "Machine learning is the core, and we are rethinking the way we do things. We are trying our best to apply it to us. In all products: search engine, advertising, YouTube or app store. We are still in the early stages, but you can see that we are applying machine learning to all places in a systematic way."

Obviously, if Google wants to apply machine learning to all its products, it requires engineers to master this completely different technology than traditional programming. As Peter Domingos wrote in The Master Algorithm, a very popular machine learning field: "Machine learning is a sunrise technology, a self-constructing technology." To write such a system you need to do the following: Work: Identify the right data, choose the right algorithm, make sure you build the right conditions for a successful run, and finally trust the system's ability to work.

First knowledge of machine learning and neural networks

Jeff Dean, Google's machine learning team leader, said: "The more people think about how to solve problems in this way, the better we will do. If every engineer has some knowledge of machine learning."

John Giannandrea has been a key player in Google Machine Learning for many years, and he has recently become the head of Google's search engine division. But when he first came to Google in 2010, he didn't know much about machine learning or neural networks. Around 2011, some news from the Neurospatial Processing System Summit (NIPS) made him feel very shocked. It seems that on the annual NIPS, there are always some teams that will announce the use of machine learning to subvert problems that have been difficult to break through, such as translation, language recognition or visual issues.

He said: "When I first heard about the NIPS summit, it was not famous, but in the past three years, the number of participants has soared. About 6,000 people participated last year."

Jeff Dean: Google Authoritative Computer Scientist

This not only promotes the development of neural network algorithms, but also brings more powerful computing power from the Moore's Law effect, as well as the exponential growth of data obtained from the massive user behavior of companies such as Google and Facebook. Machine learning The new era of continuous improvement has begun. Like some people, Giannandrea believes that machine learning is at the heart of the company.

Google's enthusiasm for machine learning not only means a change in programming technology, but also a solemn commitment to technology, and promises to give computers unprecedented power. This technology is inspired by the structure of the brain, and its forefront is to build deep learning algorithms around complex neural networks. Google's brain is Google's attempt in deep learning, and Google's artificial intelligence company DeepMind, which was bought in January 2014 for $500 million, also focuses on deep learning. The AlphaGo system, which defeated the World Champion of Go, was designed by DeepMind, which raised concerns about intelligent robots and killing robots.

For those who hold the attitude that "artificial intelligence will kill us," Giannandrea thinks they don't understand the situation. Giannandrea believes that machine learning systems will be disruptive both in the medical field and in the automotive driving arena. Although machine learning does not replace humans, it will change humanity.

Giannandrea exemplifies the power of machine learning, and Google Photos can position images that users point to. Giannandrea said: "When people first experience this product, they think that something different is happening, because the computer is no longer just calculating the recommended content for you, or suggesting what video you watch. In fact, the computer Understanding the content in the picture. This is a really new field. In some narrow areas, you can see that some people think that these learning systems have surpassed humans."

Make unimaginable products possible

It is undeniable that Google has always understood the concept of machine learning. The founder of Google is a follower of artificial intelligence. Machine learning has been incorporated into Google products, but recently Google has paid more attention to neural networks.

In fact, Google has taught engineers an in-house course in machine learning for more than a decade. In early 2005, Peter Norvig, who was in charge of the search business, advised research scientist David Pablo Cohn that he should investigate whether Google can use online classes in projects organized by Carnegie Mellon University. Cohn's conclusion is that only Google can teach such an internal classroom because Google operates far more than other institutions. Therefore, Norvig booked a large room on the 43rd floor, a two-hour meeting on Wednesdays, and even Jeff Dean participated in several times. Cohn said: "That's the best class in the world. They are all better engineers than me!" The course is on fire, and the situation is even out of control. People in the Bangalore office in India can't make a reservation call until the middle of the night. A few years later, some Googlers made a short video of these courses together and no longer had live meetings. Cohn thinks this may be the predecessor of MOOC. In the next few years, Google also conducted several other machine learning training attempts, but these attempts were lacking in order and continuity. Before Cohn left Google in 2010, he said that machine learning suddenly became Google's top priority.

Gradually, these engineers' efforts in machine learning are increasingly reflected in Google's popular products. Since machine learning mainly covers areas such as vision, speech, voice recognition and translation, machine learning has undoubtedly become a key part of voice search, translation, photos and so on. Jeff Dean, co-founder of disruptive systems such as Big Table and MapReduce, said: "Before, we might use machine learning in several subcomponents of a system. Now, we are actually replacing the entire system with machine learning. Rather than building a better machine learning model for each part."

At the same time, machine learning has made possible the previously unimaginable product features. For example, in November last year, the smart reply function of Gmail was released. This began with a conversation between Greg Corrado, co-founder of the Google Brain Project, and Gmail engineer Bálint Miklós. Corrado previously worked with the Gmail team to study the use of machine learning algorithms in spam detection and message classification, but Miklós proposed a more radical approach. If the team can invent a way to automatically generate a response to a message through machine learning, it can save the mobile user from the hassle of typing a reply on the keypad.

Google has kept Corrado working closely with his own team and the Gmail team, which adds to the possibilities of implementation. Corrado said: "Machine learning is both a science and an art. Just like cooking, although it involves chemistry, but to do something really interesting, you have to know how to combine the ingredients in front of you."

Traditional artificial intelligence methods for language understanding depend on the language rules embedded in the system, but in this project, the system uses self-learning data.

However, the key to making smart responses truly viable is that success can be easily defined, that is, to achieve a reasonable response like a real-life mail.

However, when the team started testing smart responses, the user noticed a strange thing: the system often made some untimely romantic responses. Corrado said: "It tends to reply 'I love you' at any time when I don't know how to reply. This is not a software vulnerability, but we mistakenly let it do it. The program has learned human behavior to some extent. The subtle side. For example, when you have nowhere to go, saying 'I love you' will be a good defense strategy."

The smart reply that was released last November was a success. Now users of the Gmail Inbox app don't have to type a word themselves, just select one of the three recommended emails to complete the reply. The content of these referrals is often unreasonable. One-tenth of the responses sent by Inbox users of mobile phones are created by machine learning systems.

The turning point of machine learning

In a series of examples where Google proves the efficiency of machine learning, smart reply is just one of them. But perhaps when machine learning becomes an important part of the search business, perhaps the turning point will finally come. The search business is Google's flagship product, a vault that occupies almost all of Google's revenue. To some extent, search has always been based on artificial intelligence. But for many years, Google's most precious algorithms have brought us 'ten blue links' (the ten blue links that respond to Google search requests), which is destined to be the top priority of machine learning algorithms.

To some extent, this is a cultural boycott. The prestigious search guru Amit Singhal was an assistant to the legendary computer scientist Gerald Salton. Salton's groundbreaking work in document retrieval inspired Singhal, who helped modify the graduate code of Brin and Page into a program that could be extended for use in today's Internet age. He has come up with amazing results from these 20th-century methods, and some suspect that he has brought some learners into this complex system equivalent to Google's lifeline.

At the beginning of 2014, Google’s machine learning guru thought it needed to be changed. The experiments envisioned by the Dean team later proved to be critical to the ability to search: how well a file in the sort matches the query request.

The results are useful, and the system is now part of the search, called Rank Brain, which went live in April 2015. Google still upholds its past features and is vague about how the system can improve search performance, but Dean said: "Rank Brain has been integrated into a large number of queries" and affected the actual ranking. In addition, the system works well. Among the hundreds of signals used by Google Search to calculate rankings, Rank Brain ranked third in practicality.

Google’s new challenge is to transform its engineering team to familiarize everyone with machine learning. This is the goal that many other companies are pursuing now, especially Facebook, which is a giant in machine learning and deep learning, just like Google. The recruitment competition for fresh graduates in this field is fierce, and Google seeks to maintain its early leadership. Over the years, there has been a joke in the academic world: even if it is not needed, Google is hiring the top students, and this is just In order to avoid them going to competitors. Domingos said: "My students, no matter who they are, will always receive Google's offer." Now, the competition has become more intense. Just last week, Google announced that it would open a new machine learning research lab in Zurich, where there will be a lot of jobs to fill.

Train more engineers for future artificial intelligence

Machine learning requires different ways of thinking. People become masters of programming usually because they grew up by implementing complete control over the programming system. Machine learning also requires a certain amount of mathematics and statistics, and this is what many programmers, even those who can compress the program to a surprising length, never bother to learn.

This also requires a considerable degree of patience. Robson said: "The machine learning model is not a static code. You have to constantly feed it with data. We constantly update the model and learn, add more data, and, for example, adjust the way we make predictions about our future. It makes people feel like a living, breathing thing. This is different types of engineering development."

Giannandrea added: "In fact, this is a discipline that uses different algorithms for experimentation, or a discipline that studies which training data can produce better results in your use cases. The computer science part will not go far. But There will be more attention to mathematics and statistics, and less attention will be paid to writing 500,000 lines of code."

Dean said: "At the end of the training day, the mathematics used in these models is no longer so complicated. For most of the engineers we hire at Google, this is achievable."

To further help the growing team of machine learning experts, Google has created a powerful set of tools to help engineers select the right model when training algorithms and accelerate the process of training and refining. The most powerful of these tools is TensorFlow, a system that accelerates the neural network building process. TensorFLow was developed by Google's brain team, and Dean and his colleague Rajat Monga participated in the development of it; it can make machine learning popular by standardizing the often tedious and esoteric details of system development.

Although this altruistic behavior spread to the artificial intelligence community has plagued Google, it also acknowledges that a new generation of programmers familiar with its internal machine learning tools is quite good for Google's recruitment. Still, the features of TensorFlow, coupled with Google's approval, quickly became a favorite in the machine learning programming community. Giannandrea said that when Google offered its first online TensorFlow course, 75 million people signed up.

Google still keeps a lot of good things for its own programmers. Internally, the company has a toolbox that can be unparalleled to complement machine learning, the Tensor Processing Unit (TPU). The TPU is a microprocessor chip optimized for running machine learning language programs, just as a graphics processing unit (GPU) is designed to accelerate the single-pixel calculation of on-screen pixels. There are probably tens of thousands of TPUs in the company's giant data center servers. By giving its neural network computing power, TPU has already given Google a huge advantage.

But because Google's biggest demand is for people who design and improve these systems, just as Google is in the midst of improving its software training tools, the company is also frantically polishing its experiments with training machine learning engineers.

There are other smaller jobs that are also attracting outsiders to Google's machine learning. Earlier this spring, Google launched the Brain Residency project, which was designed to bring promising outsiders into the Google brain team for a year of intensive training. Although some of the 27 machine learning students from different disciplines in the initial project may eventually stay in Google, the purpose of the training is to release them into the wild and use their super powers to spread Google across the data globe. Machine learning version.

So in a sense, in a world where machine learning is taking center of the stage, Google, with artificial intelligence as its center, has plans to maintain its dominant position, and Carson Holgate learned this in her ninja course. The plan is centered.

Eyeglass Frames

Eyeglass frames are an important part of eyewear, mainly supporting the lenses and providing an aesthetic effect. The materials used are mainly metal, plastic or resin, natural materials and so on. A fine eyeglass frame should be made of a stable, safe, reliable and skin-friendly material that is light, strong and does not deform.
We custom etching eyeglass frames with drawings provided by customers. The raw material we more use for spectacle frame is stainless steel. The thickness we can achieve is 0.03-1.0mm. We are equipped with professional metal etching equipment and exposure development equipment. Our customized metal eyeglass frames can achieve burr-free finishes, no notch.

Etching Eyeglass Frames,Spectacle Frame,Customized Metal Eyeglass Frames,Chemical Etching Eyeglass Frames

SHAOXING HUALI ELECTRONICS CO., LTD. , https://www.cnsxhuali.com