Improving Out-of-Distribution Detection with Markov Logic Networks

Out-of-distribution (OOD) detection is essential for ensuring the reliability of deep learning models operating in open-world scenarios. Current OOD detectors mainly rely on statistical models to identify unusual patterns in the latent representations of a deep neural network. This work proposes to augment existing OOD detectors with probabilistic reasoning, utilizing Markov logic networks (MLNs). MLNs connect first-order logic with probabilistic reasoning to assign probabilities to inputs based on weighted logical constraints defined over human-understandable concepts, which offers improved explainability. Through extensive experiments on multiple datasets, we demonstrate that MLNs can significantly enhance the performance of a wide range of existing OOD detectors while maintaining computational efficiency. Furthermore, we introduce a simple algorithm for learning logical constraints for OOD detection from a dataset and showcase its effectiveness.

Modern AI systems can confidently misinterpret inputs they’ve never seen before, like a blue stop sign, posing risks in real-world applications. Typical detectors look for odd patterns in a model’s internal signals, but these methods can miss obvious semantic errors and offer no clear reason why something is flagged. We propose adding a probabilistic “reasoning layer” based on simple human-readable rules (e.g., “stop signs are red octagons”) to existing detectors. We also introduce an efficient algorithm that automatically learns these rules from examples. On benchmarks like traffic signs and face attributes, our hybrid approach consistently improves detection rates and pinpoints which rules were broken. By combining pattern-based detection with rule-driven reasoning, this work makes AI systems more reliable and transparent.