Reproduce “A Semantic Loss Function for Deep Learning with Symbolic Knowledge” 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
Jingyi Xu, Zilu Zhang, Tal Friedman, Yitao Liang, Guy Van den Broeck
The authors introduce semantic loss, which measures whether neural network Boolean predictions satisfy a logical constraint. It gives differentiable feedback about allowed combinations, not individual labels alone. On a generated 4-by-4 grid-path task, it raised coherent accuracy from 5.62% to 28.51% and constraint accuracy from 6.99% to 69.89%, while edge accuracy fell slightly; valid paths need not be shortest.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.