Recently, Professor Liu Sannyuya and his team from the Faculty of Artificial Intelligence in Education achieved a significant breakthrough by developing a novel method to uncover learning patterns. This approach reveals the underlying principles of cognitive skill acquisition from large-scale, naturally occurring behavioral data, offering an efficient and promising tool for educational science research. Their work, titled “Automated discovery of symbolic laws governing skill acquisition from naturally occurring data,” was published online in Nature Computational Science, a sub-journal of Nature.

Cognitive skill acquisition is a key research area in both educational and cognitive sciences. Understanding how skills are acquired can help people learn more effectively and efficiently. However, because skill acquisition involves complex cognitive and psychological processes, the patterns uncovered by traditional experimental methods are often debated and lack broad applicability. This study embraces the AI4Science approach, developing innovative intelligent algorithms to automatically uncover the underlying principles of skill acquisition from large-scale, naturally occurring log data. To overcome challenges like hidden learning states and the vast complexity of operator search, the study introduces a two-stage pattern discovery algorithm. First, a deep learning model is trained using an autoregressive approach to estimate learning states and assess the importance of various features. Next, a symbolic distillation technique translates the neural network model into algebraic equations, allowing for an explicit and interpretable expression of the learning principles.

Simulation results show that this method can accurately identify key variables and reconstruct the preset skill acquisition equations within a certain range of noise, validating the effectiveness of the proposed approach. When applied to large-scale real-world cognitive skill training data, the skill acquisition patterns discovered by this method significantly outperform traditional and mainstream learning models in terms of fitness and goodness of fit. Additionally, the method uncovered two new types of cognitive skill acquisition patterns—logarithmic rate and inverse power rate—and confirmed several findings from previous studies.

In recent years, Professor Liu Sannyuya and his research team have centered their efforts on the national strategic priority of “Artificial Intelligence + Education” and the latest advances in the field. They have systematically conducted innovative research on computational theories, methods, and applications in education. Their work has been published in top-tier journals and conferences across education and information sciences, including Educational Research, C&E, ACM TOIS, IEEE TKDE, TNNLS, TEVC, TII, as well as AAAI, AIED, WWW, and ACM MM. They have developed the AI4EduSci paradigm for intelligent educational science and pioneered new directions in computational education, making significant contributions to building a uniquely Chinese educational science knowledge system and advancing the nation’s goal of strengthening its education sector.

Journal Information: Nature Computational Science is a premier international academic journal published by the renowned Nature family. It focuses on innovative applications of computational science to uncover new insights, tackle complex real-world challenges, and advance multidisciplinary research through the interdisciplinary use of emerging computational technologies. The journal features high-quality research spanning computational methods, data science, artificial intelligence, AI4Science, and other related fields within computational science.

Paper Information: Central China Normal University is the sole institution responsible for this paper. Professor Yang Zongkai is the corresponding author, with Professor Liu Sannyuya as the first author. Lecturer Shen Xiaoxuan and Professor Sun Jianwen also serve as co-corresponding authors. This research was funded by major projects, including the National Natural Science Foundation’s “Theoretical and Key Technology Research on AI Empowering Teaching and Learning” and the Ministry of Science and Technology’s Next-Generation AI National Major Project, “Research on Learner Cognition and Affective Computing for Smart Education.”

Paper Links：https://www.nature.com/articles/s43588-024-00629-0