On the afternoon of May 26, 2024, at the invitation of Hong Kong Graduate School of Business, Dr. Sun Dongning, researcher and doctoral supervisor of Pengcheng Laboratory's Mathematical Science and Interdisciplinary Frontier Department, visited "HKGSB Financial Observation" and brought the audience an online wonderful sharing on the theme of "the development process and key paradigm shifts of AI and large models".

The lecture was streamed live online, with a total of 2,021 views and 1729 likes.

Dr. Dongning Sun holds a master's degree from Peking University, a PhD in Biomedical engineering from Columbia University, and a post-doctoral degree in Biomedical Engineering from Johns Hopkins University. He has successively worked in Swiss Re Capital Management and Consulting Company, Citibank Investment Bank, Swiss Bank Investment Bank, Deutsche Bank Investment Bank and engaged in investment and market making business in the US. In 2014, he was appointed Chief Investment Officer of Derivatives and quantitative Investments of Ping An Fund.

Dr. Sun Dongning is currently a researcher in the Department of Mathematical Science and Cross Frontiers of Pengcheng Laboratory, doctoral supervisor, Professor of Shengao Jinke Chair of the Chinese University of Hong Kong, Standing Committee member of the Digital Finance Branch of China Computer Society, Executive director of China Quantitative Investment Club, member of the Financial Technology Branch of China Society of Industrial and Applied Mathematics, etc.

In the lecture, Dr. Sun Dongning delved into the nature of artificial intelligence (AI), defining it as machine intelligence driven by data and algorithms. He pointed out that the core of AI is pattern recognition, which is learned and corrected by intelligent programs to realize the mapping from input to output.

Dr. Sun reviewed three peaks in the development of AI, including the early search technology and logical reasoning, to the rise of expert systems, and the rise of deep learning, big data, and computing power. He emphasized the methodological shift from logical reasoning to statistical inference, and the goal shift from general AI to functional AI.

Dr. Sun then discussed in detail the hierarchical feature description capabilities of deep learning models and the technical advantages of direct input of raw data to reduce the stress of feature engineering. Dr. Dongning Sun introduced the principle of neural network modeling to describe the nonlinear behavior of systems, and explained the application and limitations of recurrent neural networks (RNN) and their variants in describing time series.

Dr. Dongning Sun points to two core technology paradigm shifts that have driven the explosion of large models in recent years. The Transformer model family and the self-attention mechanism bring the ability to process long sequences in parallel and capture long distance dependencies in sequence data. The transition from supervised learning to self-supervised learning provides a rich corpus for model training using a wide range of unlabeled data.

With the advent of LLM (Large Language Models) era, model capabilities emerge with scale, that is, "Bigger is different this time".

Through a vivid example, Dr. Sun Dongning described intelligence as the compression process from a complex high-dimensional world to a low-dimensional structure, emphasizing the importance of massive corpus data compression and the understanding of language structure and logic for the "thinking ability" of LLM.

Dr. Sun Dongning deeply analyzed the application of artificial intelligence in the field of financial technology, especially the application of time-series quantitative large models in financial data prediction and optimization.

Dr. Sun also discussed the challenges of financial timing models, including market efficiency, market style variability, reflexivity and interpretability, and stressed the importance of financial security and stability and risk monitoring and prevention and control. This paper introduces the results of the "financial wind tunnel", which is based on market data and adopts the digital twin method of mathematical modeling and quantitative scenario analysis for risk monitoring and simulation.

Finally, Dr. Sun Dongning introduced Pengcheng Lab's big brain model and its ongoing research and development and ecological construction in vertical fields such as government affairs, education, manufacturing, finance and medical care.

Dr. Sun Dongning's sharing not only provided deep insights on AI and large models, but also demonstrated the practical application and broad prospects for future development of these technologies in the financial field.

In the interactive session, the audience actively participated and actively asked questions on the opportunities and challenges of big models in quantitative finance, the integration of big models with cutting-edge technologies such as quantum computing and brain-computer interface, and the integration prospect of big models with emerging machine learning paradigms such as meta-learning and autonomous reasoning. Dr. Sun Dongning conducted in-depth analysis and detailed answers.

This lecture is the first holding of the "HKGSB Financial Observation", and will continue to be carried out in the future.

"HKGSB Financial Observation" is initiated and hosted by Hong Kong Business School. It is committed to building a professional sharing platform with content, quality and depth, covering hot issues in the economic and financial fields. In the form of online live broadcasting, senior experts are invited to share the latest developments in professional fields, express original views, and communicate with the audience in depth.