Reproduce “Tratto: A Neuro-Symbolic Approach to Deriving Axiomatic Test Oracles” 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
Davide Molinelli, Alberto Martin-Lopez, Elliott Zackrone, Beyza Eken, Michael D. Ernst, Mauro Pezzè
Tratto, a neuro-symbolic system developed by researchers from Constructor Institute, USI, and University of Washington, automates the derivation of axiomatic test oracles for software functions. It integrates neural network generalization with symbolic reasoning precision, achieving a 73% accuracy in oracle generation and outperforming state-of-the-art symbolic and pure neural methods.
Key insights
- The neuro-symbolic integration effectively combines the generalization capabilities of neural networks with the precision and correctness guarantees of symbolic reasoning.
- An iterative, token-by-token generation strategy, guided by strict grammatical and contextual constraints, is critical for producing syntactically and semantically valid and compilable axiomatic oracles.
- A multitask learning approach within the neural module, encompassing both oracle evaluation and token selection, contributes to enhanced overall oracle generation accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.