Reproduce “Deep Adaptive Semantic Logic (DASL): Compiling Declarative Knowledge into Deep Neural 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
Karan Sikka, Andrew Silberfarb, John Byrnes, Indranil Sur, Ed Chow, Ajay Divakaran, Richard Rohwer
SRI International developed Deep Adaptive Semantic Logic (DASL), a framework that compiles First-Order Logic declarative knowledge directly into deep neural networks for end-to-end training. The approach significantly reduces labeled data requirements, achieving 97.7% accuracy on MNIST with a ~1000-fold data reduction and improving visual relationship detection performance by up to 10.7% in data-scarce and zero-shot conditions.
Key insights
- Full First-Order Logic can be formally grounded and compiled into a differentiable neural network, allowing for highly expressive knowledge integration.
- Using a logit transformation for truth values effectively addresses the vanishing gradient problem in continuous logical operations, enabling deeper symbolic reasoning.
- Declarative knowledge can serve as a powerful regularizer, significantly reducing the dependence on labeled data and improving model generalization, especially in data-scarce environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.