Reproduce “Think-Verify-Revise: Neuro-Symbolic Visual Reasoning with Vision-Language Models and Dynamic Logic Tensor Networks” 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
Homayoun Afshari, Pietro Basci, Alessandro Russo, Lia Morra
The authors present a system that proposes Sudoku rules from images and tests them with a differentiable verifier. A vision-language model uses three examples; feedback guides revisions. It recovered rules for MNIST, EMNIST, and KMNIST, but not FMNIST within 20 iterations. On EMNIST, mean test ROC AUC was 0.93 ± 0.10, versus 0.65 ± 0.14 for LTN-IND C results.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.