CNCC2023 will be held in Shenyang from October 26 to 28, during which 129 technical forums will be held, covering more than 30 directions such as artificial intelligence, security, computing +, software engineering, education, network, chip, cloud computing, etc. This paper introduces the technical forum "Prospective Problems and Challenges of Large Language Models in the Field of Mathematics: Theory, Methods and Applications" to be held on October 28.
Mathematics has always been seen as a litmus test for artificial intelligence. When the big language model breaks through its "innate defects" and successfully solves the challenges of mathematics, the world's artificial intelligence will enter a new era. The forum will deeply discuss the progress and challenges of large language model reasoning ability in the field of mathematics, analyze how to improve the mathematical reasoning ability of large language model, and look forward to the prospect and research direction of large language model in the field of mathematics.
To register and learn more about the technical forum, please identify the QR code below to enter the official website of CNCC2023.
Large language models that perform amazingly in generating text passages, simulating human dialogue, and solving mathematical problems are one of the hottest areas of artificial intelligence development in recent years. The emergence of the large-language model instance ChatGPT has injected a shot in the arm into its development, bringing a new paradigm of AI R&D and application and changing the AI ecological pattern. However, large models perform poorly in complex reasoning represented by mathematical reasoning, and it is difficult to cope with research problems in the field of mathematics. On the other hand, due to the shortcomings of large models in mathematical concept understanding and generation, interpretability and transparency, generalization ability, error and stability, personalization and interestingness, it is difficult to meet the application requirements of basic education.
How to improve the mathematical ability of large models and break through the congenital shortcomings of language models has become the focus of attention in the field of artificial intelligence. This forum will start from the past and present of large models, combined with the latest research results, and analyze and discuss the progress and challenges of large language model reasoning ability in the field of mathematics. including how to accurately measure the mathematical reasoning performance of language models; The reasons for the high error rate of the model in mathematical reasoning and the possible solution strategies were discussed. Discuss new methods and technologies to make the model produce more stable and coherent problem-solving steps in mathematical reasoning; Discuss how to improve the learning experience and effect of the model; In the future, how the application of language models in mathematical reasoning will change the traditional teaching model.
Forum arrangement
| order | topic | Keynote speaker | unit |
| 1 | Large model base empowerment: practice and thinking from general to specialized | Wu Fei | Zhejiang University |
| 2 | Large-scale multimodal pre-training models for teaching | Liu Jun | Xi'an Jiaotong University |
| 3 | Large models make personalized learning at scale truly possible | Tian Mi | Good Future Group |
| 4 | A conversational big model of fusion educational psychology: EduChat | Zhou Aimin | East China Normal University |
Chairman of the Forum
Liu Zitao
Professor of Jinan University and Dean of Guangdong Institute of Smart Education
Senior member of CCF, professor of Guangdong Institute of Smart Education, Jinan University, doctoral supervisor, dean of the institute. He has published more than 80 papers in top conferences and journals in the field of artificial intelligence such as ICML and NeurIPS, and has authorized more than 40 invention patents at home and abroad. He serves as an executive member of the International Association for Artificial Intelligence Education and the program chairman of the 25th International Conference on Artificial Intelligence Education (AIED). Presided over and participated in many scientific research works such as the national key research and development plan, scientific and technological innovation 2030-new generation artificial intelligence major projects, etc. The research results have been reported by many news media and news websites, including CCTV News Network, CCTV Zhengdian Finance, People's Daily News, Xinhuanet and so on.
Co-Chairs
Wang Yanfeng
Assistant Director of Shanghai Artificial Intelligence Laboratory
Deputy Dean of the Institute of Artificial Intelligence of Shanghai Jiao Tong University and Deputy Director of the Collaborative Innovation Center of Future Media Network
He is currently a deputy to the 16th Shanghai Municipal People's Congress. Member of the National Science and Technology Innovation 2030 - "New Generation Artificial Intelligence" Major Project Guide Compilation Expert Group, Member of the National Science and Technology Innovation 2030 - "New Generation Artificial Intelligence" Major Project Management Expert Group, Member of the "Internet +" Action Expert Advisory Committee of the National Development and Reform Commission, and Member of the National Development and Reform Commission's Artificial Intelligence Industry Advisory Expert Committee.
Forum speaker
Wu Fei
Professor of Zhejiang University and Director of the Institute of Artificial Intelligence
Senior member of CCF, Qiushi Distinguished Professor of Zhejiang University, doctoral supervisor. His main research areas are artificial intelligence, multimedia analysis and retrieval. Executive Deputy Dean of Shanghai Advanced Research Institute of Zhejiang University and Director of Institute of Artificial Intelligence of Zhejiang University. Winner of the National Science Foundation for Outstanding Young Scholars (2016), member of the major project management expert group of science and technology innovation 2030 "new generation artificial intelligence" of the Ministry of Science and Technology and science and technology and preparation expert, leader of the working group of the artificial intelligence science and technology innovation expert group of the Ministry of Education (2018.8-2020.12), executive editor of the information and electronic engineering discipline of the journal of the Chinese Academy of Engineering "Engineering", winner of the 9th Yongping Outstanding Teaching Contribution Award of Zhejiang University. He has won the first prize of scientific and technological progress of the Ministry of Education (ranked first) and the first prize of scientific and technological progress of the Chinese Institute of Electronics (ranked first), authored "Introduction to Artificial Intelligence: Models and Algorithms" and "Into Artificial Intelligence" (Higher Education Press), and opened the first batch of national online first-class courses "Artificial Intelligence: Models and Algorithms".
Large model base empowerment: practice and thinking from general to specialized
This report introduces representative algorithms such as pre-training, supervised fine-tuning and human-loop feedback in the training process of large models, and depicts the characteristics of data and models as large and language points in the machine learning paradigm of "pre-trained model + prompt learning + prediction". At the same time, it introduces the vertical field large model Zhihai Sanle for the teaching of the core course "Introduction to Artificial Intelligence" of the 101 Project built with high-quality textbook-level corpus, and the vertical field large model for intelligent justice created through logic enrichment corpus, Zhihai-Record.
Liu Jun
Professor/PhD supervisor, School of Computer Science, Xi'an Jiaotong University
He is a senior member of CCF, a senior visiting scholar at Stanford University, the director of Shaanxi Key Laboratory of Big Data Knowledge Engineering, and has been selected as a national leading talent. He serves on the editorial board of IEEE TNNLS and guest editor of several journals. His main research interests are natural language processing, computer vision, and intelligent education. In recent years, it has undertaken the National Key Research and Development Program and the National Natural Science Foundation of China. He has published more than 100 papers in IEEE TPAMI, IJCV, IEEE TKDE, 2 monographs, and authorized 20 invention patents; He won the second prize of National Science and Technology Progress Award, the second prize of National Teaching Achievement Award, the first prize of Shaanxi Province Technological Invention Award, and the special prize of Science and Technology Progress Award of Chinese Society of Automation; He has won honors and awards such as Shaanxi Youbo Instructor and Wang Kuancheng Education Award.
Large-scale multimodal pre-training models for teaching
The large model represented by ChatGPT has greatly improved the general intelligence and content generation capabilities, and is expected to become an artificial intelligence infrastructure for the digitalization of education. However, the application of large models to teaching scenarios still faces a series of challenges: first, there are problems such as illusions, ideologies, and values in generating content; Second, the ability to understand cross-media teaching content is weak, especially for teaching resources containing schematic diagrams; Third, it lacks the ability to divide and conquer planning and reasoning; Fourth, it is difficult to generate personalized content that meets the needs and cognitive level of users. In this regard, it is necessary to build a large-scale multi-modal pre-training item model for education, break through the key technology of "one foot at the door", and promote the autonomy and control of key links.
Tian Mi
COTO of Good Future Group
CCF member, CTO of Good Future Group. Fully responsible for future technology research and development. Currently, he focuses on the research and development of educational large models, leading the team to develop MathGPT large models and AI Tutor products based on large models. He was a senior technical director of Alibaba Group and a director of R&D for Tencent Group. Tian Mi holds a bachelor's and master's degree from the School of Computer Science of Beihang University, and his master's degree is under the tutelage of Academician Li Deyi, whose research direction is artificial intelligence and data mining.
Large models make personalized learning at scale truly possible
Teaching according to aptitude and teaching without distinction has been the goal pursued by mankind for thousands of years. The rise of LLM technology has made it possible to solve large-scale personalized learning. The essence of LLM is a more efficient way to learn from data and apply it. With the blessing of AI capabilities, the new learning method of "students' self-learning + AI Q&A" will become a wide range of possibilities. Unfortunately, although there are many excellent LLMs around the world, such as GPT-4, they cannot be directly used for personalized learning, especially mathematics. In response to this problem, Xueersi launched MathGPT, the first 100-billion-level large model in the field of mathematics in China, creatively combining the capabilities of large models and computing engines, and effectively solving the three major challenges of LLM in the field of mathematics: solving problems, clarifying steps, and interesting and vivid content. This report will elaborate on how MathGPT solves the application problems in the field of mathematics from the aspects of the development logic and effect of MathGPT.
Zhou Aimin
Dean, School of Computer Science and Technology, East China Normal University
CCF member, Dean of School of Computer Science and Technology, East China Normal University, Dean of Shanghai Institute of Intelligent Education of East China Normal University, Deputy Director of Intelligent Education Laboratory of East China Normal University of the Ministry of Education, and Head of Shanghai IV Peak Discipline (Intelligent Education). His main research areas include evolutionary optimization and learning, machine learning, and intelligent education. He has published more than 40 academic papers in SCI Region I/CCF Class A journal conferences, and related achievements have been cited more than 8400 times by Google Scholar. He serves as associate editor of the journal Swarm and Evolutionary Computation, editorial board member of Complex & Intelligent Systems, and associate editor of the Chinese Journal of Electronics. He won the title of Elsevier's 2020-2022 China Highly Cited Scholar and the second prize of Shanghai Higher Education Excellent Teaching Achievement (Undergraduate Education) in 2022.
A conversational big model of fusion educational psychology: EduChat
Large models show good processing ability in general tasks, but still face many problems in the vertical field of education, such as hallucinations and knowledge update lag, lack of heuristic guidance, lack of deep emotional interaction, lack of personalization, etc., which seriously affect the application of large models. In response to these problems, the team recently developed EduChat, a dialogue model for education verticals, under the guidance of theories such as education and psychology. This report will introduce some of our explorations in educational big models, including details such as educational function design, data construction, model training, and applications in mathematics and psychology.
This year marks CNCC's 20th anniversary. Over the past two decades, CNCC has gradually developed to cover 129 technical forums in dozens of directions, with more than 700 domestic and foreign speakers actively participating and more than 13,000 people registering as an annual event in the field of computing. Twenty years of continuous surpassing, as an annual event with many participants, large scale and high level in the field of domestic computing, CCF will carefully plan to bring participants a cutting-edge collision and look forward to the future technology feast, so that every participant can enhance their professional value and gain momentum in the super large professional platform of CNCC! Wait for you to come, act now, welcome to participate and register!