Research map / Large language models / Retrieval and knowledge
Recurrent Relational Networks
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 |
|---|---|---|---|
| use recurrent relational message passing instead of competing differentiable Sudoku methods | Sudoku solver · Sudoku (17 givens) | Accuracy: 92.5% -> 96.6% | Helped |
| use 1 reasoning step instead of 3 (ablation) | RRN · bAbI (20 tasks) | Mean Error (%): 0.46 ± 0.77 -> 0.48 ± 0.79 | No clear effect |
| use 2 reasoning steps instead of 3 (ablation) | RRN · bAbI (20 tasks) | Mean Error (%): 0.46 ± 0.77 -> 0.46 ± 0.76 | No clear effect |
| replace the 4-layer MLP encoding with a single linear layer (ablation) | RRN · bAbI (20 tasks) | Mean Error (%): 0.46 ± 0.77 -> 0.20 ± 0.01 | Helped |
| encode questions together with fact representations (ablation) | RRN · bAbI (20 tasks) | Mean Error (%): 0.53 ± 0.91 -> 0.46 ± 0.77 | Helped |
| add dropout (ablation) | RRN · bAbI (20 tasks) | Mean Error (%): 1.74 ± 1.28 -> 0.46 ± 0.77 | Helped |
| remove row and column position features (ablation) | Sudoku RRN · Sudoku (17 givens, 64 steps) | Accuracy: 96.6% -> 96.7% | No clear effect |
| train with loss only at the last step instead of at every step (ablation) | Sudoku RRN · Sudoku | Accuracy | Hurt |
| run the trained network for 64 rather than 32 steps | Sudoku RRN · Sudoku (17 givens) | Accuracy: 94.1% -> 96.6% | Helped |
| use multiple recurrent reasoning steps instead of a single step (ablation) | Relational Network · Sudoku (17 givens) | Accuracy: 0% -> 96.6% | Helped |
| use recurrent relational reasoning for multi-jump questions rather than a single-step relational network | Pretty-CLEVR relational reasoning model · Pretty-CLEVR | Accuracy | Helped |
| increase the number of recurrent reasoning steps to match the required relational chain | RRN · Pretty-CLEVR | Accuracy | Helped |
Closest papers by meaning
- A simple neural network module for relational reasoning (NeurIPS 2017 · similarity 0.91)
- Relational recurrent neural networks (NeurIPS 2018 · similarity 0.89)
- When No Paths Lead to Rome: Benchmarking Systematic Neural Relational Reasoning (NeurIPS 2025 · similarity 0.87)
- Reasoning of Large Language Models over Knowledge Graphs with Super-Relations (ICLR 2025 · similarity 0.87)
- RelGNN: Composite Message Passing for Relational Deep Learning (ICML 2025 · similarity 0.86)
- Systematic Relational Reasoning With Epistemic Graph Neural Networks (ICLR 2025 · similarity 0.86)
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.