<ul><li>Study investigates generalization behavior of neural architectures (Transformers, Graph Convolution Networks, LSTMs) using propositional logic.</li><li>Models were tested on generating satisfying assignments for logical formulas, emphasizing structured and interpretable settings.</li><li>While all models performed well in-distribution, generalization to unseen operator combinations, especially negation, remained challenging.</li><li>Findings suggest persistent limitations in standard architectures' ability to learn systematic representations of logical operators, indicating a need for stronger inductive biases.</li></ul>

Propositional Logic for Probing Generalization in Neural Networks

Discover more