Title: From End-to-end to Step-by-step: Integrating Machine Learning and Logical Reasoning through Abductive Learning

Speaker: Wangzhou Dai (Nanjing University)

Time: 2025/3/27 (Thursday) 15:10-18:00

Location: Room B112, Lee Shau Kee Humanities Buildings No.2 (李兆基人文学苑2号楼), Peking University

Abstract:

Integrating machine learning with logical reasoning is a foundational challenge in artificial intelligence, particularly in bridging the gap between statistical patterns in data and explicit symbolic representations. This talk provides an accessible introduction to philosophers and logicians who seek to understand how contemporary machine learning methods can be harmonized with traditional forms of formal reasoning. Beginning with an overview of induction and abduction as distinct yet complementary forms of inference, we will delve deeply into the concept of abductive learning—a framework combining the statistical strengths of machine learning with the explanatory power of logical reasoning. A key philosophical and computational problem, the emergence of symbolic representations from sub-symbolic perception, will be explored to demonstrate how machines may autonomously identify meaningful concepts without explicit human-defined categories. Finally, we discuss the transition from "end-to-end" learning systems, which behave like opaque statistical black-boxes, toward interpretable, step-by-step decision-making architectures inspired by human cognitive processes. This shift holds promise not only for improving the robustness and explainability of AI systems but also for advancing our philosophical understanding of cognition, logic, and the nature of symbols themselves.

References:

- W.-Z. Dai, Q. Xu, Y. Yu, and Z.-H. Zhou. Bridging machine learning and logical reasoning by abductive learning. In: Advances in Neural Information Processing Systems 32 (NeurIPS’19), Vancouver, Canada, 2019.
- W.-Z. Dai and S. H. Muggleton, Abductive Knowledge Induction From Raw Data, In: Proceedings of the 30th International Joint Conference on Artificial Intelligence (IJCAI’21), pp. 1845-1851.
- W.-C. Hu, W.-Z. Dai, Y. Jiang, Z.-H. Zhou, Efficient Rectification of Neuro-Symbolic Reasoning Inconsistencies by Abductive Reflection, In: Proceedings of the 39th Annual AAAI Conference on Artificial Intelligence (AAAI’25), Philadelphia, PA, 2025. (Outstanding Paper Award)
- L. Tao, Y.-X. Huang, W.-Z. Dai and Y. Jiang. Deciphering Raw Data in Neuro-Symbolic Learning with Provable Guarantees. In: Proceedings of the 38th Annual AAAI Conference on Artificial Intelligence (AAAI’24), Vancouver, Canada, 2024.
- Y.-X. Huang, W.-Z. Dai, L.-W. Cai, S. H. Muggleton and Y. Jiang. Fast Abductive Learning by Similarity-based Consistency Optimization, In: Proceedings of the 35th Conference on Neural Information Processing Systems (NeurIPS’21), pp.26574-26584.