At present, artificial intelligence has been considered as the technological high point of the next era, a variety of AI related technologies and products continue to emerge, enterprises in all fields are also actively embrace the darling of this era, in order to rely on artificial intelligence to break through business pain points, improve work efficiency, and achieve enterprise transformation and upgrading. The three most active technology directions in Fintech in recent years have been artificial intelligence, big data and blockchain. Among them, artificial intelligence technology is the core, which has been applied in quantitative investment, risk identification, credit assessment, intelligent customer service and other fields. By some measures, the financial system is one of the areas where ai can be deeply integrated and produce value. The financial field has been fully data-oriented, providing sufficient data raw materials for the application of artificial intelligence; At the same time, the boundaries of various branches and fields within the financial system are relatively clear, enabling the application of artificial intelligence in vertical fields. However, artificial intelligence + financial, after all, still in the early stages of the development in the technology practice, fall to the ground and AI liquid products, and artificial intelligence in the application of Internet and electronic commerce also has the difference, and how Fintech embedded in financial business, to improve investment efficiency, strengthen risk control and implementation technology have been exploring, There are not only objective reasons such as business scenarios, data accumulation and technology precipitation, but also subjective factors such as cognitive differences and conceptual hype. This paper discusses how to make artificial intelligence + finance more grounded, respectively from the perspective of concept, technology, algorithm, scene and so on.
1, eliminate the false and preserve the true, distinguish priorities and priorities
The technical fields of artificial intelligence mainly include image recognition, speech recognition, natural language processing, knowledge engineering, etc., and the application fields include question answering robot, autonomous driving, intelligent security, recommendation system, computational advertising, quantitative investment, etc. The industry background, data basis and personnel allocation are all different, which also causes some confusion. Many participants and participating companies have different ai professional capabilities and practical experience. They often confuse the concept and label conventional technology as AI, misleading and confusing the public. As a result, there are also many artificial intelligence counterfeit products, which ultimately affect the terminal experience and have a bad impact on the subsequent promotion of related products. There is also a similar phenomenon in the financial field. For all kinds of landed and unlanded products, first of all, they are good at “exaggerating” in publicity and promotion, which emphasizes showing off skills rather than functions. Or “graft”, directly copy the Internet companies or robot field of some mature applications or hot technology. This makes people have a better understanding of artificial intelligence than expected, and may “win praise” from the public opinion, but this is not healthy. It is essentially creating “obstacles” for the landing and in-depth application of artificial intelligence in the financial vertical field. Too much amplification of the halo of cutting-edge technology, weaken the understanding of the nature of the business, all kinds of “smart packaging” on the stage, a blind, disturb the pace of reason. Some cutting-edge AI technology products may not be really suitable for business scenarios, resulting in a lot of capital, manpower, time and other costs of high investment, low return.
Perhaps, the promotion of any cutting-edge technology can not get rid of the negative of the negative law, I believe that with the development of financial technology and financial business transformation, the in-depth application of artificial intelligence in the financial industry, the general trend is steady forward. Only relevant practitioners in this process to eliminate the false and the true, distinguish priorities. Identify technical capabilities, understand business scenarios, but also inclusive of misunderstanding, the gradual integration of artificial intelligence technology into the business, improve efficiency, solve pain points, so that artificial intelligence + finance less devious.
2，Data construction and accumulation
Now the rapid development of artificial intelligence is based on big data, it can be said that “there is no data without AI”. Data is the fuel of artificial intelligence, such as data understanding, business modeling, factor selection, model optimization, portrait system, etc., all need the support of basic data, how to lay a good data foundation, the accurate and complete data collection, its importance is self-evident. At present, many financial companies are trying to achieve business transformation and performance overtaking by taking advantage of artificial intelligence, but in the actual landing process, it is impossible to avoid three data problems: first, lack of data; Second, lack of good data; Third, lack of higher-order data. And these three “lack” have essential difference.
The first “lack” often occurs in more traditional business scenarios, and the introduction of artificial intelligence method, even a lot of business process data has not been unified collection and management, such as some financial audit business, a lot of data is still in the form of bills, pictures or paper documents, etc. How to digitize these “paper” data and participate in the unified data construction is the first problem to be solved. For another example, public opinion data on social networks can also affect the volatility of financial markets. In this scenario, it is also necessary to collect social network data.
The second “deficiency” generally occurs when the application of artificial intelligence in a certain business scenario has been completed and the business model has been determined. In order to ensure and improve the effect, it is more urgent to see whether the data is good or not. For example, for the risk assessment model based on Logistic regression, the accuracy and consistency of the data related to risk factors are very important. Data cleaning and validation work is almost the key to this application scenario.
The third “missing” usually occurs the application of artificial intelligence in a business scenario is relatively mature, also develop faster, on the basis of the comprehensive and high quality data, has the demand for higher order data, quantitative investment, for example, with the widespread application of the deep study on quantitative investment, the traditional factors is no guarantee that the “robustness” higher investment model output, A variety of derived data, knowledge base, relational data, and abstract factors have become a new engine to continuously promote and challenge quantitative investment, which is as important as the optimization of learning models such as deep learning, reinforcement learning and transfer learning.
With the continuous development of big data and application of artificial intelligence, especially for deep learning, the boundaries between “data” and “knowledge” is becoming more and more fuzzy, so on the basis of the traditional data warehouse, explore how to from the “data construction” to “knowledge base construction”, for the development of the application of artificial intelligence and machine learning has profound meaning.
3,Supervised learning has mixed achievements
Currently, the vast majority of AI applications rely on supervised machine learning techniques: annotated samples, training models, target prediction. It seems to have become the machine learning practitioners and artificial intelligence novice’s routine “battle plan”, more people will think that this is the artificial intelligence “routine”. Indeed, with the development of deep learning in the past few years, supervised learning has achieved great success. The algorithm’s ability to represent data has been continuously strengthened, and various optimization methods have also continuously improved the robustness of the model. In spite of this, there are still many unavoidable defects in the practical application of supervised learning: Sample first of all, the labeling of access is becoming more and more is not easy, especially in the financial information of NLP applications, based on the artificial corpus annotation is a “huge” project, must have a large corpus, takes a lot of artificial tagging, and on the depth of labels to the definition of the business, financial corpus will need a mark according to the different business scenarios. Secondly, due to the weak reasoning ability of supervised learning, it can perform very well on the trained projects, but the untrained samples cannot be identified, even in deep learning. The factors influencing the financial market are complex and changeable, and the randomness and uncertainty can be seen everywhere. It is very difficult to obtain a stable prediction model with low variance and low deviation by relying on supervised learning.
4, real-time architecture, better
With the rise of Internet finance, many traditional financial businesses are now facing many unprecedented challenges. Mobile and big data have become the most obvious points of competition. How to quickly acquire data and process and make decisions based on real-time machine learning technology has become the forefront of competition. The traditional mode of offline financial data analysis, low-frequency processing and regular manual decision analysis will greatly reduce the efficiency of getting the “truth”. Financial market data change rapidly, and “lag” is the biggest risk. In this context, more and more attention is paid to real-time machine learning.
Real-time machine learning can be divided into three categories: hard real-time, that is, the response system can respond to the request immediately after receiving the request, and make a response and feedback, which requires very high timeliness; Soft real-time means that the response system begins to process the response immediately after receiving the request and gives feedback within a short time. The timeliness is less than that of hard real-time. Batch real-time means to process all the data in a certain time window. In this architecture, there is usually a distributed queue to buffer and process the data in the time window. At present, offline statistical learning and data mining based on Matlab, R language and Python can still meet the needs of some basic business scenarios, but artificial intelligence algorithm technology boosted by big data and parallel computing will increasingly have a strong dependence on high-performance computing architecture, even fusion. Only in this way can we better realize function landing and differentiation competition.
Most modern real-time machine learning and data processing architectures can be summarized as Lambda architectures, an architectural paradigm capable of processing large amounts of data in real time, designed with response delays, processing fluxes, and high fault tolerance in mind. In order to meet the requirements of real-time applications, Lambda architecture can be divided into three parts: Serving Layer, StreamingLayer, and Batch Layer. The main elements of the real-time response layer include databases (such as Redis, Druid, etc.) that support fast update read and write, and algorithm services that can perform fast read response. The fast processing layer is composed of various streaming processing platforms, such as Apache Spark, Storm, etc. Batch processing mainly includes various offline processing tools, such as MySQL and Hive.
In the financial sector, the design of AI products requires both in-depth research into the business and an in-depth understanding of all ai capabilities, which requires quantitative analysis and research of a practical problem. On the basis of in-depth investigation and understanding of the object information, simplified assumptions are made, internal laws are analyzed, and appropriate artificial intelligence technology is used for adaptation. The cognitive architecture of, algorithms, data, business will also continuously improve: first of all, is the understanding of the business scenarios and data must be well done, return to nature of the business, not misled by the “package” products available in the market, and affect the objective evaluation of problem, completes the data cleaning and data warehouse construction, explore the knowledge base construction; Second, the field of artificial intelligence technology, supervision of learning is not all of artificial intelligence, planning, reasoning and perception in all aspects, such as continuous development, we should broaden the mind, to think in more quantification to think + financial, thinking about how to make artificial intelligence AI based on econometric and classic financial engineering theory and methods to serve the business nature, reduce risk, Improve efficiency; Thirdly, the development of artificial intelligence not only depends on complex logic and formulas, but also depends on a few open source communities. How to customize real-time machine learning architecture and build high-performance computing platform are equally important to the implementation of artificial intelligence. In particular, the scenarios that require high real-time computing architecture, such as online learning, dynamic programming, risk identification, real-time decision-making, etc.
With any technological revolution, there is a cognitive and habitual shift that brings pain and unease, from social software that has decimated entrenched texting and phone calls to e-commerce that has decimated brick-and-mortar retail overnight. The penetration and development of AI in the financial industry will have much impact on our current financial industry. It is not yet clear, but AI is strongly boosted by the continuous development of the Internet and big data, and driven by the general trend of modern production and life from digital to intelligent. Only every practitioner of financial ecology is compatible, pragmatic, innovative, brave to think and try, so that artificial intelligence + finance is more and more grounded, and will surely hear thunder in the silent place.