#author("2021-06-12T17:06:33+09:00","","")
[[TopPage]]

Earlier on December 17, Microsoft announced on its official blog a data set of 100000 questions and answers that researchers can use to create systems that can read and answer questions like humans. In addition, Microsoft plans to follow the example of Imagenet, cooperate with others, and finally launch a formal competition. This data set is called Ms Marco, which means Microsoft machine reading comprehension. The team behind it claims that this is the most useful data set in this category at present, because this data set is based on anonymous real data. By opening the data set to more researchers for free, the team hopes to promote breakthroughs in machine reading, just as previous researchers have made disruptive breakthroughs in image recognition and speech recognition. They also hope that this opening up will promote the realization of the long-term goal of "AgI / artificial general intelligence", that is, to create machines that can think like human beings. "In order to achieve the goal of AI, we first need machines to be human like," said rangan Majumder, program manager of partner group in Bing search engine division of Microsoft, who led the projectClass to read and understand documents. This data set is a step in that direction. " Majumder said the system for answering complex questions is still in its infancy. Search engines like Bing and virtual assistants like Xiaona can only answer some basic questions, such as "when does Hanukkah begin?" Or "how much is 2000 times 43?" But in many cases, search engines and virtual assistants only direct users to some search results, Majumder said. Of course, users will still get the information they want, but that also requires users to find the answer link they need in the search results list. In order to achieve a better automatic question answering system, researchers need more powerful training data. Such training data needs to be able to teach AI systems to identify problems and organize answers, and ultimately build their own answers based on specific problems they have never seen before. Majumder and his team (including some Microsoft researchers and product developers) said that the MS Marco dataset is very useful because its problems are based on real, anonymous queries from Bing search engine and Xiaona virtual assistant. The team chose these questions based on what the researchers thought were more interesting queries. In addition, the answers to these questions are written manually based on real web pages, and the accuracy has been verified. By providing real questions and answers, these researchers say they can train to be betterA system that responds to the nuances and complexities of questions that people often ask, including those that have no clear answers or multiple possible answers. For example, this dataset contains the question: "what foods did ancient Greeks eat?" To answer this question correctly, they need to retrieve information from multiple documents, and finally give food such as grains, cakes, milk, olives, fish, garlic and cabbage as answers. Deng Li, chief AI scientist of Microsoft and partner research manager of deep learning technology center, said that previous datasets had some specific limitations and limitations in design. This makes it easier for researchers to create solutions that can be formalized as so-called "classification problem" by machine learning researchers, but it can not help the machine understand the actual text of the problem. Deng Li said that Ms Marco was designed to help researchers experiment with more advanced deep learning models, so as to promote the further development of artificial intelligence research. "Our data set is not just to use real-world data, but also to remove these restrictions so that a new generation of deep learning models can figure out numbers before they answer questions," he saidAccording to Majumder said that the system's ability to answer complex questions can help people acquire information more effectively, thereby enhancing human capabilities. Let's take an example. Suppose a Canadian student needs to know whether she is eligible for a loan program. The search engine may direct the user to a series of related websites, and then she needs to read those articles herself before she can draw a conclusion. But if she has better tools, her virtual assistant can help her scan the information and give a more detailed and even personalized answer. "Given that much of the world's knowledge exists in the form of writing, if we can make machines read and understand documents like humans, we open the door to all kinds of possibilities," Majumder said Long term goal: "artificial general intelligence"    at least for now, researchers are still far from being able to create a system that can understand what people say, see or write - many people call it "artificial general intelligence". In the past few years, Microsoft and other machine learning and artificial intelligence researchers have made great progress in creating systems to recognize words in conversations, as well as in accurately recognizing the composition of images. "Microsoft has already played a leading role in speech recognition and image recognition, and now we are going to lead the way in reading comprehension," Majumder saidResearch. " However, he mentioned that this is not a problem that any single company can solve. Majumder said one of the reasons his team opened up the data set was that they wanted to work with others in the field. MS Marco is similar to the training set in other fields of machine learning and artificial intelligence, including Imagenet data set, which is considered to be the first data set to test the progress of image recognition. A research team of Microsoft used Imagenet to test its first deep residual network, which has greatly improved the accuracy of image recognition. The MS Marco team also plans to follow the example of Imagenet and create a team ranking with the best research results. In the end, they may create a more formal competition like Imagenet's annual challenge. Any researcher who wants to download and use it for non-commercial applications can use the MS Marco dataset for free.   link to this article:  Science and technology media network - a new media platform dedicated to promoting the development of innovative technology and focusing on science and technology news communication. More interesting content, please pay attention to WeChat official account: gdkjcm (Editor: Summer noise) 
http://pandora.nla.gov.au/external.html?link=https://site-4440904-9831-7590.mystrikingly.com/blog/360-wins-microsoft-msrc
http://sc.devb.gov.hk/TuniS/www.spotoclub.com/product-category/aws/
http://ezproxy.cityu.edu.hk/login?url=http://paginasempreto.blogspot.com/2021/03/certificacoes-microsoft-spoto.html?m=1
https://intensedebate.com/people/melodycopy1
http://www.authorstream.com/dinghyfox1/


トップ   新規 一覧 検索 最終更新   ヘルプ   最終更新のRSS