Article Preview
Top1. Introduction
The digital economy industry has become the most representative high-tech industry, deeply integrated with other industries, plays a significant role in improving industrial efficiency and promoting new industries, and has become a new driving force for China's economic transformation and upgrading. Countries around the world have regarded the development of digital economy as an important part of economic growth and improving international competitiveness (Kitouni 2018).
Today's world has entered an important period in which information technology, represented by the Internet, has penetrated into the field of traditional economy and promoted the transformation of various economies to digital economy. The innovation of business model and the change of economic format in digital economy have greatly reduced the economic cost and greatly improved the efficiency of International trade, which has a significant impact on the division of global value chain. With the enhancement of China's economic strength, the deepening of financial reform and the improvement of the market (Yeh nd), RMB will not only increase greatly in the scale of overseas circulation, but also continuously enrich its functions, and gradually transition from sovereign currency to international currency.
With the development of computer technology, all kinds of scientific industries have achieved unprecedented development. Long M thinks that domain adaptation generalizes learning models across source and target domains, which are sampled from different distributions. In order to enhance the invariance of depth representation and make it more transferable in various fields, he proposed a unified depth adaptation framework for joint learning of transitive representation and classifier, so as to make full use of the advantages of depth learning and optimal two sample matching to achieve scalable domain adaptation. The framework includes two interdependent paradigms, namely, unsupervised pre training using depth de-noising automatic encoder for effective training of depth model, and supervised fine-tuning using depth neural network for effective utilization of discrimination information. Both of them learn by embedding depth representation into regenerative kernel. Although his research is tenable in theory, there is no concrete experiment to prove it (Long 2016). Charalampous K proposed an unsupervised online deep learning algorithm for video sequence action recognition. Each computing node of his example forms a cluster and calculates the point representation (Lv 2020). Then, the first-order transformation matrix stores and continuously updates the continuous transformations between clusters. Viterbi algorithm deals with both spatial and temporal information. Although the algorithm he used has good performance, there are errors in the calculation results (Charalampous 2016). Zhang L believes that in the past few decades, people have developed machine learning tools, such as quantitative structure-activity relationship (QSAR) modeling, which can quickly and cheaply identify potential bioactive molecules from millions of candidate compounds. He summarized the history of machine learning and provided insights on the recently developed deep learning methods and their applications in rational drug discovery. Although his research is more comprehensive, many definitions are not accurate (Zhang 2017). Mahmud M believes that the improvement of computing power, along with the increase of data storage speed and the decrease of computing cost, has enabled scientists in various fields to apply these technologies to data sets that were difficult to process due to scale and complexity. He will conduct a comprehensive investigation on the application of DL, RL and deep RL technology in biological data mining (Feng 2020). In addition, he compared the performance of applying DL technology to different datasets in various application domains. Finally, he outlined the problems to be solved in this challenging research area and discussed the future development prospects. Although his research is comprehensive, there is a lack of experimental data (Mahmud 2017).