Research map / Large language models / Reasoning and chain of thought
Neural Algorithmic Reasoning Without Intermediate Supervision
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 |
|---|---|---|---|
| replace standard no-hint mode with encoded-decoded no-hint mode, no position scalars, fixed step count | Triplet-GMPNN no-hint · CLRS sorting | micro-F1: 68.5 -> 90.7 | Helped |
| add contrastive regularisation to no-hint (ours) on sorting | Triplet-GMPNN no-hint (ours) · CLRS Insertion sort | micro-F1: 90.67 ± 8.22 -> 98.74 ± 1.12 | Helped |
| add contrastive term on MST-Kruskal | no-hint (ours) · CLRS MST-Kruskal | micro-F1: 92.89 ± 0.92 -> 93.15 ± 1.01 | No clear effect |
| add contrastive term on MST-Prim | no-hint (ours) · CLRS MST-Prim | micro-F1: 85.70 ± 1.44 -> 85.20 ± 1.77 | No clear effect |
| add contrastive term on Binary search | no-hint (ours) · CLRS Binary search | micro-F1: 85.29 ± 4.52 -> 87.12 ± 2.23 | Mixed |
| add contrastive term on Minimum | no-hint (ours) · CLRS Minimum | micro-F1: 99.98 ± 0.02 -> 99.99 ± 0.01 | No clear effect |
| no-hint (ours) versus hint-supervised baseline | Triplet-GMPNN · CLRS 18 tasks | micro-F1: 14 out of 18 tasks | Helped |
| no-hint (ours) versus standard no-hint on Strongly connected components | no-hint · CLRS SCC | micro-F1: 57.63 ± 0.68 -> 66.38 ± 7.12 | Helped |
| no-hint (ours) versus standard no-hint on Binary search | no-hint · CLRS Binary search | micro-F1: 93.21 ± 1.10 -> 85.29 ± 4.52 | Hurt |
| no-hint (ours) versus standard no-hint on Floyd-Warshall | no-hint · CLRS Floyd-Warshall | micro-F1: 40.80 ± 2.90 -> 42.16 ± 2.14 | No clear effect |
| no-hint (ours) versus Hint-ReLIC on Heapsort | no-hint (ours) · CLRS Heapsort | micro-F1: 95.16 ± 1.27 -> 90.67 ± 8.22 | Hurt |
| step count linear -> quadratic for sorting (ablation) | no-hint (ours) · CLRS sorting | micro-F1: 90.67 ± 8.22 -> 90.97 ± 8.03 | No clear effect |
Closest papers by meaning
- Discrete Neural Algorithmic Reasoning (ICML 2025 · similarity 0.89)
- Open-Book Neural Algorithmic Reasoning (NeurIPS 2024 · similarity 0.89)
- Primal-Dual Neural Algorithmic Reasoning (ICML 2025 · similarity 0.87)
- Deep Equilibrium Algorithmic Reasoning (NeurIPS 2024 · similarity 0.86)
- On the Markov Property of Neural Algorithmic Reasoning: Analyses and Methods (ICLR 2024 · similarity 0.86)
- Right for the Right Reasons: Avoiding Reasoning Shortcuts via Prototypical Neurosymbolic AI (NeurIPS 2025 · similarity 0.84)
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.