The role of force data algorithm in artificial intelligence
AI instry chain can be divided into basic layer, technology layer and application layer. The basic layer is divided again according to computing power, data and algorithm, which plays a supporting role in the overall superstructure; The technology layer is divided into computer vision, intelligent speech, natural language processing and so on, which is the most attractive part of AI; The application layer customizes and develops exclusive services according to the needs of different scenarios, which is a way for AI to truly enable the instry
at present, the commercialization of artificial intelligence has basically reached a stage of maturity in terms of computing power, algorithm and technology. If we want to be more practical and solve the specific pain points of the instry, we need a large number of labeled data for algorithm training support. It can be said that the data determines the landing degree of AI.
there are many applications of classification in real life, such as spam classification, for example, judging the patient's illness
for example, guessing whether it will rain tomorrow
to make any choice, you can learn this problem-solving model from historical data
relevant data show that by 2025, the amount of data generated will be as high as 163zb, of which 90% are unstructured data. These unstructured data can only be awakened after cleaning and labeling, which leads to a continuous demand for data cleaning and labeling.
some emerging technologies have also achieved good results in the field of knowledge discovery, such as neural network and decision tree. With enough data and computing power, they can automatically complete many valuable functions without human care
data mining is to use the algorithms and technologies of statistics and artificial intelligence technology to encapsulate these advanced and complex technologies, so that people can complete the same functions without mastering these technologies, and focus more on the problems they want to solve
the main difference between data mining and the two lies in the adaptability of the algorithm to the large amount of data. The algorithm of data mining must face the data set with more than 100000 records and have good performance; Periodic data set updating data mining needs to consider how to deal with these incremental data without calculating from scratch; data mining also needs to consider how to deal with the problem that the data set is larger than the memory and parallel processing; in addition, data mining is oriented to solve engineering problems.
first, deep learning is an algorithm revolution, which brings together many algorithms scattered in various fields. Fu Sheng said that in the past, he found a problem in the process of interviewing people and looking at the company, that is, there is a huge algorithm gap between different majors, such as voice, image and driverless. But after the emergence of deep learning, it can not only solve the problem of voice, but also solve the problem of image and driverless. So once there is such a normalized algorithm, the most intelligent brain power of human beings will be concentrated here
Fu Sheng believes that with the technology of deep learning, the biggest panic should be the big companies, because they have accumulated a lot of technology, but these technology accumulation has been subverted under the impact of deep learning. For example, not long ago, Fu Sheng met with the head of a laboratory of a well-known large domestic company. He told Fu Sheng that he had worked in the field of translation for seven years. Later, when he saw a paper by Google, he suddenly found that his original technology had accumulated in vain. So Fu Sheng thinks that deep learning is essentially recing technical barriers. The more big companies want to do, the more they panic
the second point is that algorithm driven has become data driven. Fu Sheng said that although each paper can promote the whole instry, the driving force of the algorithm has been greatly reced e to the immobilization of the basic algorithm model. Judging from today's trend, instrialization and data-driven are the main ones. The huge amount of data generated after a large amount of advance, may be far better than a paper. A large number of data also need to be annotated. So large scale annotation data has become the core competitiveness. What do you mean? For example, the sound of people talking is sent to the Internet, which is not the data available for AI. You have to find someone to mark out the key points in the sound data, which is the effective data available for AI
Data tagging has just begun. Some companies look very big, but their tagging data is very small. Today, there are a lot of data on the Internet for people to use freely. The key path here is to figure out how to label these data and how to quickly generate your own data set. For example, for alphago, the chess score in human history is far from enough for it to learn. The latest alphago has to remove the chess score of human beings, because it thinks that human beings are not playing well. The proct manager of Google translation said that the reason why the latest version of Google translation has been greatly improved is that it has captured a large amount of data from the Internet and concted secondary annotation. However, a lot of data on the Internet is translated by Google before, and they have to slowly eliminate the poor data
thirdly, Fu Sheng said that the opportunity of deep learning lies in the combination with application rather than technology output. Because in the future, deep learning will become the basic technology application, and many companies will have the research and development ability of deep learning, it is difficult to imagine that a company can succeed only by providing technology output
as the next prospective outlet, many enterprises have begun to enter the field of artificial intelligence, but as the infrastructure, deep learning still has a long way to go. Wsjfabric, a big data processing information service provider, believes that the key to deep learning lies in the processing of big data. Taking alphago as an example, the first contact with go is of course based on the human chess score. It is because after studying a large number of human chess score that it is possible to defeat human beings. Therefore, the cooperation between AI enterprises and big data enterprises is imperative. On the one hand, it focuses on data collection; On the other hand, the algorithm is developed to process the data and realize deep learning. For big data enterprises, through cooperation and sharing of data research results, they graally position themselves as the driving force of social progress and realize the transformation from it to DT.
The relationship between artificial intelligence and big data is very close. In fact, the development of big data promotes the development of artificial intelligence technology to a great extent, because data is one of the three foundations of artificial intelligence technology (the other two foundations are algorithm and computing power). From the current technical architecture of artificial intelligence, the current artificial intelligence is still highly dependent on data. It can also be said that there is no intelligence without data
to understand the relationship between artificial intelligence and big data, we can describe it through machine learning. On the one hand, machine learning is an important part of artificial intelligence technology, on the other hand, machine learning is also widely used in the field of big data, so machine learning can be seen as a bridge between artificial intelligence and big data
machine learning has five major steps, including data collection, algorithm design, algorithm implementation, algorithm training and algorithm verification. The machine learning algorithm that has completed the verification can be applied in the actual scene. Through the steps of machine learning, we can find that data collection is the basis of machine learning. Without data collection, algorithm training and algorithm verification cannot be completed. In fact, data also has a very direct impact on algorithm design. From this point of view, before the research and development of artificial intelligence, there must be data first
at present, machine learning is not only widely used in the field of artificial intelligence, but also one of the two common ways of big data analysis, so many practitioners in the big data instry can smoothly turn to the field of artificial intelligence through machine learning, which also blurs the technical boundary between big data and artificial intelligence to a certain extent. In fact, many enterprises engaged in artificial intelligence research and development have a certain big data base, which is one of the reasons why many Internet enterprises can walk in the forefront of artificial intelligence research and development
finally, the development of big data and artificial intelligence needs two important foundations, namely, the Internet of things and cloud computing. The Internet of things not only provides the main data source channel for big data, but also provides scene support for the landing application of artificial intelligence procts, while cloud computing provides computing power support for big data and artificial intelligence. Therefore, to engage in the research and development of big data and artificial intelligence, we also need to master certain knowledge of Internet of things and cloud computing
