Reproduce “Program Induction by Rationale Generation : Learning to Solve and Explain Algebraic Word Problems” 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
Wang Ling, Dani Yogatama, Chris Dyer, Phil Blunsom
Researchers at DeepMind and the University of Oxford introduce a method for solving algebraic word problems by generating a natural language rationale that serves as an intermediate guide for inducing latent arithmetic programs. This approach achieves 36.4% accuracy on a new 100,000-example dataset, AQuA, while also producing step-by-step explanations for the solutions.
Key insights
- Generating human-readable rationales acts as effective scaffolding for inducing complex latent programs, making an otherwise intractable program induction task tractable.
- Integrating rationale generation into the problem-solving process inherently provides interpretability and transparency, aligning with Explainable AI principles by showing the derivation steps.
- The approach effectively bridges neural network capabilities (sequence-to-sequence generation) with symbolic reasoning (pre-defined mathematical operations and program execution).
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.