Reproduce “Quantitative Linear Logic for Neuro-Symbolic Learning and Verification” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Thomas Flinkow, Ekaterina Komendantskaya, Matteo Capucci, Rosemary Monahan
The authors introduce Quantitative Linear Logic (QLL), which turns logical rules on additive quantities such as logits into differentiable training losses using smooth connectives. They test MNIST robustness, requiring the true-class logit to remain largest within an ε = 0.2 neighborhood. QLL reaches 46.0 ± 23.2% verified satisfaction, versus 10.4 ± 21.6% for DL2; small-model experiments are demonstrative, not evidence of real-world scaling.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.