According to the Wired magazine website, on the west coast of Australia, biologist Amanda Hodgson controlled drones to fly high above the Indian Ocean. The dugong expert used drones to help them observe the endangered subjects. However, Hodgson and her team did not have the ability to screen all the photos taken. Finding dugongs in 45,000 photos is too difficult for untrained eyes. Her solution is to give this work to the deep neural network to complete.
The neural network is a machine learning model, and the "face recognition" that the public is familiar with is one of its applications. In addition, the reason why the mobile phone's intelligent voice assistant can understand what you are saying, Google search engine can present accurate search results, it also has its credit. By simulating human brain neural networks, these scanning mathematical models acquire specialized functions through analysis of large amounts of data. Dr. Hodgson used this technique to find traces of dugong in thousands of aerial photographs. Hodgson's neural network is based on TensorFlow, the second generation of artificial intelligence learning system developed by Google.
Because dugongs are used to preying below the surface of the water, the task of detecting these animals requires exceptional precision. "Theirs are easily confused with the glare of the water," she said. Now her neural network can identify dugongs that spread 80% of the sea surface.
The project is still in its early stages, but it demonstrates the wide-ranging impact of deep learning in the past year. Deep learning is infinite in 2016. This ancient technology has been given new energy to help Google defeat humanity in the world-famous Go war. This was almost impossible a few months ago. AlphaGo ("Go" in English) is just the most prominent example. In the past year, deep learning is no longer just a niche for tech geeks, but it has turned to the scenery. Google, Facebook, Microsoft and Amazon are taking a new look from the inside out. In turn, the help of these Internet giants – through open source and cloud services – has also accelerated the popularity of deep learning.
New translationLast year, neural networks took image recognition technology to a new level in applications such as Google Photo. Google Now and Microsoft Xiaona also achieved superior speech recognition because of their blessing. This year, it was the turn of the translation community to be transformed. Machine translation has achieved a great leap forward. In September, Google launched the "Neural Machine Translation" service. This translation runs entirely through the neural network, reducing the translation error rate by 55% to 85%.
Google trains neural networks through aggregated data from a large number of existing translations. The training materials include both poor translations from the old translation software and Shinjada translations provided by human language experts, which undoubtedly add to the quality of the material. Deep learning has the magic of overcoming defects: despite the uneven quality of training materials, neural networks can eventually achieve far lower levels of translation.
Although Mike Schuster, Google's chief engineer, is frank that their creations are far from perfect, it is still a breakthrough achievement. Since the service is based entirely on deep learning, future improvements will be much easier. Developers can focus on making improvements to the system as a whole, rather than tangling widgets as they have in the past.
In addition to Google, Microsoft is also working in the same direction. This month, Microsoft also released a new version of its translation application. It claims to be able to translate instantly between nine languages. Microsoft's vice president, Harry Shum, said Microsoft's translation system is also fully operational on neural networks. This means that the level of Microsoft translation has also increased rapidly.
New chatIn 2016, deep learning also played a major role in the field of chat bots. One of the most compelling is Google Allo. Launched this fall, Allo provides instant, intelligent responses through analysis of user text and photos. The implementation of its features is based on Google's previous technology called "Smart Reply," which is largely similar to email technology.
Allo is more than just a chat app, it can also enhance your Google search experience without your awareness. The program helps the search engine understand your needs so that the results returned by the search are more relevant to your needs. According to David Orr, Google's search product manager, the program can't answer without deep learning. “Using neural networks is the only way we find it,†he said. “We must use the most advanced technology we have at hand.â€
Despite its strengths, the real dialogue still leaves the neural network incapable. There is still a long way to go to create such a "chat robot" that is completely fake. At the moment, researchers at Google, Facebook, and other places are actively exploring deep learning techniques to achieve ambitious goals one day. What is certain is that these technological explorations will bring about as great progress as "speech recognition," "image recognition," and "machine translation." "Chat robot" is the next technology frontier.
New data centerGoogle can't stop on the road of deep learning. This summer, after creating the famous AlphaGo, Google DeepMind Labs leader Demis Hassabis said they also developed an AI to manage Google's global computer data center network. Using a technology called Deep Reinforcement Learning, AI intelligently manages the switching and temperature control of cooling fans in the server. In short, more than 120 functions in a data center are controlled by it.
Bloomberg reported that the deployment of the AI ​​helped Google save hundreds of millions of dollars. In 2014, Google spent $650 million to acquire DeepMind, and now it has fully recovered the cost. Currently DeepMind is planning to install more sensors outside of these computing facilities and collect more data to train AI to a higher level.
New cloud computingWhen Internet giants are armed with new technology, they also donate it to the public through their own services. At the end of 2015, Google announced the open source of TensorFlow. In just one year, this once proprietary software has benefited millions of people like Amanda Hodgson. At the same time, Google is also working with Microsoft and Amazon to provide their own deep learning technology in cloud computing services, allowing any individual or organization developer to use them to build their own programs. "Artificial Intelligence Services" may become the biggest business of these three network giants.
In the past 12 months, the heat of technology has made the talents in the field hot. Fei-Fei Li is a technology leader in AI research, and Google hired her to manage her AI cloud computing organization. Amazon hired Carnegie Mellon University professor Alex Smolna to sit on his cloud computing empire. The technology giants do their utmost to recruit talents and do not give each other. Fortunately, the research results of their competition will be used by the public. This is not a bad thing.
As AI evolves, the role of computer scientists is changing. People who can write code in the traditional sense are less important, and new trends require more people who can train neural networks. The latter requires more skills than before, and it is more like inducing data to produce results than when developing something. These big companies like Google are not only actively recruiting new talents, but also guiding existing employees in this direction. The future AI will revolutionize the technology in everyone's life.
Edge Sealing Machine,Automatic Sealing Machine,Heat Sealing Machine,Automatic Edge Sealing Machine
Dongguan Yuantong Technology Co., Ltd. , https://www.ytbagmachine.com