We present a neurosymbolic reasoning and learning methodology based on a modular integration of Answer Set Programming (ASP) with Energy Based Models. Key contributions include:
-
Supporting joint optimization in continuous latent space through explicit ASP-based declarative semantics that fully incorporate background knowledge, constraints, and non-monotonic inference.
-
Advancing recent works at the interface of answer sets, probabilistic logic, and answer set modulo theories by providing a generalized model and practical platform for robust, end-to-end ASP-centric training applicable to dynamic domains (e.g., involving perception and interaction).
We provide a practical implementation, demonstrate basic use and application (with MNIST), and evaluate against the visual question-answering benchmark Clevr and the multi-object tracking benchmark MOT.
Blogger's Review: This paper combines neural networks with symbolic reasoning, showcasing the potential for optimizing reasoning and learning in dynamic environments. It offers new perspectives and tools for designing intelligent systems, particularly improving model performance on complex tasks. The incorporation of ASP allows better utilization of background knowledge, pushing the frontiers of intelligent reasoning forward.