Research map / Learning theory and algorithms / Combinatorial and search optimization
Making Neural Programming Architectures Generalize via Recursion
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| make bubble sort recursive (partial: tail call to BUBBLESORT only) (ablation) | NPI bubble sort · Bubble Sort, length 3 | accuracy: 6.7% -> 23% | Mixed |
| full recursion (BUBBLESORT, BSTEP, LSHIFT tail-recursive) (ablation) | NPI bubble sort · Bubble Sort, lengths 3-90 | accuracy: 0% to 100% at length 90 vs non-recursive | Helped |
| recursive traces for topological sort | NPI topological sort · Topological Sort, 5-70 vertices | accuracy: 6.7% -> 100% at 5 vertices; 0% -> 100% at 70 | Helped |
| recursive traces for quicksort | NPI quicksort · Quicksort, lengths 11-70 | accuracy: 73.3% -> 100% at length 11; 0% -> 100% at length 70 | Helped |
| recursive traces for bubble sort variant | NPI bubble sort variant · Bubble Sort variant, length 8 | accuracy: 6.7% -> 100% | Helped |
| recursive traces for grade-school addition (trained on 5 one-digit sums) (ablation) | NPI addition · Grade-school addition, long sums | qualitative generalization | Helped |
| recursive traces for addition | NPI addition · Grade-school addition, up to 5000 digits | generalization | No clear effect |
Closest papers by meaning
- Neural-Symbolic Recursive Machine for Systematic Generalization (ICLR 2024 · similarity 0.88)
- Gradient-Based Program Synthesis with Neurally Interpreted Languages (ICLR 2026 · similarity 0.88)
- Improving the Universality and Learnability of Neural Programmer-Interpreters with Combinator Abstraction (ICLR 2018 · similarity 0.88)
- Programming with a Differentiable Forth Interpreter (ICML 2017 · similarity 0.88)
- Differentiable Programs with Neural Libraries (ICML 2017 · similarity 0.87)
- Data-Efficient Learning with Neural Programs (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.