The concept of "machine learning" has received much attention from the scientific and technological community since its emergence in the 1950s. In recent years, "deep learning" has gradually become a new field in machine learning research. The motivation is to establish and simulate human brain for analysis. Learning neural networks that mimic the mechanisms of the human brain to identify images, sounds, and text.

The online version of the US technology media "Connected" magazine recently published a summary of the latest developments in "machine deep learning" technology. The following is the main content of the article.

In the eyes of Quoc Le, the world is made up of a series of numbers. “A digital photo is actually a number,” he said. “If you split what people say into separate phonemes, they can also be compiled into numbers.” If you follow Quoc Le, you can Digital input into the machine, the machine can read the photos and what people say, such as Facebook can recognize your face, Google can understand what you said.

But Quoc Le wants to go further, and he hopes to develop a technology that translates the entire sentence, the entire paragraph, and various types of natural language into numbers or other vectors. With this technology, computer scientists can Let the machine also get the information people see and hear. At the same time, Quoc Le is also exploring ways to make machines understand people's opinions and emotions.

Although such technology is still in its infancy, there is still a long way to go in the future, but Quoc Le has more resources for its deployment than its peers. Quoc Le is a member of the "Google Brain" project, which focuses on the "machine deep learning" field, a form of artificial intelligence that uses machines to simulate human brain for data processing.

Quoc Le, 32, has been working on voice recognition in Google, such as the voice recognition function of the Android system and the automatic labeling of network images. Both of these tasks require the support of "deep learning" technology.

In addition to Google, Internet giants such as Facebook and Microsoft are also using "deep learning" technology, and Baidu recently publicly talked about using this technology to provide customers with more accurate advertising push services. But Quoc Le hopes to push the technology to a wider range of areas, including natural language understanding, robotics, and web search.

Quoc Le has recently developed a “deep learning” technology that identifies how different words are related on the web, and Google incorporates this technology into its “knowledge map” to help it find results. Systematize knowledge so that every keyword can get a complete knowledge system.

Once troubled

Quoc Le first came into contact with artificial intelligence in the 1990s, but it really bothered him, because the machine learning system at that time relied heavily on the manual input of engineers. Although the machine has the ability to understand to a certain extent, it needs to be more cumbersome. The operation can be completed. For example, the machine at that time could not identify the photo without adding a label.

“We studied a lot of unlabeled data,” said Quoc Le, who worked with Andrew Ng, one of the founders of the Google Brain project, at Stanford University to study artificial intelligence. “If we can find it in the future. A viable algorithm for the machine to identify unlabeled data will likely change the entire computing industry. After all, most of the data on the network (such as Facebook, Twitter and Google) is unlabeled."

This is also the goal that “deep learning” technology will hope to achieve in the future. Using tens of thousands of computers to simulate the neural network in the human brain through software, the machine can acquire learning capabilities similar to humans, such as in some cases the machine can automate learning without tagging the data.

Google's cat face recognition is actually a typical case of "deep learning" technology, but after three years of research and development, the project still has not made great progress. At the same time, most commercial deep learning systems still rely on manual monitoring. “Although the practicality of cat face recognition technology is very low,” Wu Enda said, “But I think this technology represents a direction for deep learning in the future.”

Language challenge

Another challenge that “deep learning” technology needs to face is the recognition of natural language. The human language contains a lot of subtle information, and so far the scientific community has not found a way to identify these subtle information. For example, an identical vocabulary will have different meanings in different contexts or moods. Currently, most artificial intelligence systems cannot distinguish this information. “The machine is very good at processing data, but it can't cope with language symbols,” Quoc Le said. “Because language is a highly symbolic thing.”

The key to identifying a language is to find a way to translate the symbol into a number. “At the moment we have not found a way to turn the concept of language into a mathematical structure that machines can handle,” Quoc Le said. “But with the help of the Word2Vec tool, we have made some progress in this area. Hope that our machine will be in the future. Ability to automatically identify the vast amount of information posted on the network."

“People can't monitor machine learning anytime, anywhere,” says Richard Socher, who is studying for a Ph.D. at Stanford University with Quoc Le. “We hope to combine supervised learning with unsupervised learning in the future. So that the machine can achieve many things that are currently unimaginable."

Quoc Le recently joined several colleagues at Google to publish an article on the use of machine translation in deep neural network research, which talks about the use of “regressive neural networks”, which is probably the most advanced in the field of language recognition. technology.

More powerful "Google brain"

Quoc Le said in the article that the new method they found is superior to other machine translation algorithms, but this is just an application of "deep learning". Future "deep learning" technology will also be used on the network. Answers, automatic explanations, and sentiment analysis.

In order to take full advantage of these advanced algorithms, Google will not be able to expand its "machine neural network" scale, rather than confined to the field of image and speech recognition. The founder of the concept of "deep learning", Geoff Hinton, who currently works for Google, said in the introduction of the "Google Brain" project: "Like the brain of a pigeon, although it has good vision, But no one will talk to a pigeon."

In fact, even a pigeon with a fairly small brain capacity can easily surpass the most advanced "machine neural network" (including "Google Brain") in the world, and after Han Ding joined Google, it is Declaring that the future will help Google build the world's largest "machine neural network" to conduct a more comprehensive study of "deep learning."

Modular Plug

Modular Plug,modular jack rj45,modular jack cat6,mod plugs

NINGBO UONICORE ELECTRONICS CO., LTD , https://www.uonicore.com