The integration of machine learning and logical reasoning has long been considered a holy grail problem in artificial intelligence. ABductive Learning (ABL) is a paradigm that integrates machine learning and logical reasoning in a unified framework.
To facilitate the research and application of abductive learning, a research team in LAMDA group published their toolkit on 15 December 2024 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.
ABLkit is an open-source Python toolkit for abductive learning. It provides a comprehensive framework that covers the entire abductive learning workflow, including data loading, learning model development, reasoning module construction, and bridging between learning and reasoning.
ABLkit encompasses four modules: Data, Learning, Reasoning, and Bridge. The key features of ABLkit include high flexibility, easy-to-use interface, and optimized performance. The source code is available at https://github.com/AbductiveLearning/ABLkit .Performance comparison of different methods
Experiments compare ABLkit with other neuro-symbolic approaches. By effectively integrating machine learning and logical reasoning in a balanced loop, coupled with engineering optimizations, ABLkit demonstrates superior performance in terms of predictive accuracy, training time efficiency, and memory usage.
ABLkit is a Python toolkit for abductive learning. It offers high flexibility, easy-to-use interface, and optimized performance, promising to facilitate both academic research and practical applications in this field.
DOI: 10.1007/s11704-024-40085-7