nearly 5,000 technical teams showed off their skills at the afac2024 competition, and the judges were surprised by their creativity
2024-08-31
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
from fledgling undergraduates to experienced senior engineers, 4,882 teams representing different technical schools showed off their skills and competed fiercely in the afac2024 financial intelligence innovation competition. many unknown technical people impressed the judges with their genius creativity and ideas, making them call it "unexpected."
guided by the shanghai science and technology commission, supported by the china computer federation (ccf) and the shanghai science and technology entrepreneurship center, and jointly initiated by more than 20 famous universities and companies such as peking university, fudan university, the university of hong kong, and ant group, the afac2024 financial intelligence innovation competition has recently entered the final stage. this year's competition has been upgraded on the basis of last year's competition. in addition to the challenge group, two new groups, the startup group and the enterprise group, have been added, forming a comprehensive competition system structure of algorithm competition, application competition and entrepreneurship competition.
innovative solutions flourished, and the judges called it "unexpected"
the start-up group recruits start-up teams or seed companies. participants can focus on core areas such as financial technology and digital life, and design creative development and applications by combining the agentuniverse multi-agent framework and large model technology. in the start-up group of the competition, zhiyou huanjing is an ai-driven intelligent travel companion tool that provides itinerary customization, intelligent travel companionship, video travel diary generation and other capabilities, covering a variety of travel scenarios, and can customize detailed travel plans including transportation, hotels, and attractions according to user preferences, budgets, etc. it can generate videos based on the planned itinerary, allowing users to experience the entire itinerary more intuitively in advance in the form of videos, while providing intelligent travel companions with functions such as transportation, explanations, and recommendations during the trip.
the enterprise group is for small and medium-sized enterprises in the financial technology industry. participants are required to focus on emerging directions in the financial technology industry and submit new technologies and new product implementation projects that are bright and innovative. in the enterprise group of the competition, shenqing technology and several leading securities companies have conducted in-depth joint research and development in the fields of smart investment advisor assistants and personalized content generation. through the qiankunquan agent platform, it solves the problems of illusion control of large models in the financial field, business data and knowledge access, compliance and security of generated results, end-to-end efficiency of production and research operations, and quality evaluation of generated results. it is positioned as an agent platform suitable for financial institutions, integrating large model capabilities into the "system engineering" of financial business to solve business rigid needs and provide impetus for the "digital, intensive, and refined" customer management of financial institutions.
the recruitment of members for the challenge group is mainly for developers. the contestants will design projects based on the real data in the field of financial technology provided by the competition. not only originality is required, but also "implementation" is required to actually solve the scenario problems of financial business. huang chao, assistant professor at the school of computer science at the university of hong kong, served as a judge for the "identification of contradictions and discovery of vulnerabilities in long texts of financial rules" competition in the challenge group of the competition. during the review process, the contestants' innovative solutions refreshed him. "the contestants' time is actually quite tight. after the topic is released, the contestants usually only have a few days to complete the solution, but they still have a high degree of completion within the time limit, which surprised me; what's more surprising is that the contestants this time are generally young, and the entry point of the innovative solution is beyond the judges' expectations. for example, the use of prompt word design and fine-tuning to achieve compliance and error correction issues in finance, the methods are diverse and have a great possibility of implementation."
at present, the technical route of using big models to solve financial document error correction and better achieve compliance is effective. he suggested that the contestants can enhance the generalization of the solutions in the future. "at present, these solutions can achieve very good results on a certain type of financial documents, but there are many types of documents in the financial field, such as financial reports and financial analysis. if we want to achieve better standardization, we need to better use the generalization ability of big models."
huang chao suggested that the contestants use different categories of financial data to train models, or explore extracting common features of various financial documents and passing them to the big model so that the big model can understand this knowledge and enhance the model's generalization ability. when fine-tuning the model, a richer big model can be used, fine-tuned from different dimensions, adding existing financial compliance restrictions, aligning human financial compliance cognition, and promoting the model to land in real scenarios. at the same time, at present, whether it is data generation or prompt word design, the proportion of human participation is greater. automation and intelligence are important factors for the widespread implementation of solutions. therefore, huang chao suggested that in the future, an intelligent agent approach should be adopted to deeply mine various financial documents and implement more comprehensive automation in various financial fields.
grasp the technical boundaries and find the right pmf
this year, the teams participating in the afac competition are getting younger and more diverse, including undergraduate students and senior engineers who have worked in the industry for many years and love algorithms. lu xiangtao, a second-year graduate student majoring in computer science from peking university, spends three or four hours a day doing in-depth research, immersing himself in the ocean of data to optimize algorithms and solve the extremely challenging problem of processing contradictions in financial documents. xia maojin, a 32-year-old algorithm manager, balances family, work and competitions, and is increasingly aware that in the era of large models, traditional data processing and analysis methods have become difficult to solve real problems encountered in work scenarios. only by trying new ideas and methods can we stand out in the fierce competition.
in addition to the algorithm performance indicators based on accuracy, huang chao also values the innovation, core technology, and design ideas of the participating solutions. the technical pain points and challenges of different vertical fields and scenarios are not exactly the same. contestants need to understand the characteristics of financial data and the technical pain points of financial scenarios, and combine ai knowledge to choose appropriate ai tools and technologies to solve specific financial problems. "when solving these problems, you must think of new methods."
"when exploring cutting-edge ai, vitality is very important. these contestants are in an era full of opportunities. it is very exciting that they can participate in the research of big models during their undergraduate studies." he hopes that the contestants can continue to delve deeper into the field of big models, and looks forward to seeing the analysis and prediction scenarios of financial trends and market conditions in the future when big models are applied.
fan hongwei, deputy general manager of shanghai hengsheng juyuan data services co., ltd., who served as the judge of the afac competition, said that the original intention of this competition is to encourage the use of artificial intelligence technology to help the development of the financial industry and inspire entrepreneurs to innovate and start businesses in this field. during the review process, he saw innovative solutions brought by representatives of startups from different segments, covering marketing, investment research, investment consulting, risk control, data and other aspects. he also saw the contestants discussing with each other to provide the financial industry with a full-link, full-scenario, full-product intelligent overall solution, achieving a new breakthrough in the integration of industry and technology, which well reflects the original intention of the organizers of the competition.
at present, the big model has shifted from the early technology-driven to the ecosystem-driven, and the open source big model ecosystem is also accelerating its evolution. fan hongwei also admitted that the competition also showed the difficulty of the current big model implementation, but the low procurement and strong customization of open source have brought many conveniences to entrepreneurs. he hopes that entrepreneurs can continue to make good use of open source resources and participate in the prosperity of the open source ecosystem. today, the speed of technological iteration is faster than ever before. i believe that in the near future, new intelligent technologies such as big models will promote higher-quality development of the financial industry.
the competition questions come from the industry, and the solutions should go back to the industry. cui can from ant investment department said that in order for the competition proposals to be noticed by the judges, they must have new ideas, solid technology, and pmf (product market fit) solutions. he suggested that the contestants make full use of the dividends brought by technological updates, grasp the boundaries of technology, find the right pmf, and cross the cycle to make the business solid and stable. ant group will also create cooperation with an open mind to allow outstanding start-ups in the competition to pass the "germination period" and allow technology-based small and medium-sized companies to emerge.
thepaper.cn reporter jialiluo
(this article is from the paper. for more original information, please download the "the paper" app)
