Neuro-Logic Lifelong Learning
Bowen He, Xiaoan Xu, Alper Kamil Bozkurt, Vahid Tarokh, Juncheng Dong
TL;DR
This work addresses lifelong learning in inductive logic programming (ILP) by exploiting the compositionality and reuse of logic rules across tasks in a neuro-symbolic setting. It introduces a neural-symbolic framework where a shared knowledge base $B_S$ is built from multi-depth neural logic machine predicates $H_{\mathrm{NLM}}^i(B)$ and used by task-specific modules to learn new predicates with a reduced search space $\widetilde{\mathcal{F}}_t$. Empirical results on arithmetic, tree, and graph ILP tasks and BlocksWorld relational RL show forward transfer, replay-based mitigation of forgetting, and occasional backward transfer, supporting the proposed lifelong ILP paradigm. The findings suggest a path toward scalable continual reasoning in neuro-symbolic AI and motivate future work on dynamic knowledge bases and automated rule construction.
Abstract
Solving Inductive Logic Programming (ILP) problems with neural networks is a key challenge in Neural-Symbolic Ar- tificial Intelligence (AI). While most research has focused on designing novel network architectures for individual prob- lems, less effort has been devoted to exploring new learning paradigms involving a sequence of problems. In this work, we investigate lifelong learning ILP, which leverages the com- positional and transferable nature of logic rules for efficient learning of new problems. We introduce a compositional framework, demonstrating how logic rules acquired from ear- lier tasks can be efficiently reused in subsequent ones, leading to improved scalability and performance. We formalize our approach and empirically evaluate it on sequences of tasks. Experimental results validate the feasibility and advantages of this paradigm, opening new directions for continual learn- ing in Neural-Symbolic AI.