Research map / Learning theory and algorithms / Combinatorial and search optimization
Learning Simple Algorithms from Examples
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 |
|---|---|---|---|
| add dynamic discount to Q-learning (ablation) | GRU Q-learning · Single Digit Multiplication | success rate: 0% -> 100% | Helped |
| add Watkins Q(lambda) on top of dynamic discount (ablation) | GRU Q-learning · Reverse (FF controller) | success rate: 0% -> 100% | Helped |
| add penalty term (ablation) | GRU Q-learning · Walk | success rate: 10% -> 90% | Helped |
| add penalty term (ablation) | GRU Q-learning · 2-row Addition | success rate: 60% -> 100% | Helped |
| add penalty term (ablation) | GRU Q-learning · 3-row Addition (extra curriculum) | success rate: 80% -> 80% | No clear effect |
| add penalty term (ablation) | GRU Q-learning · Reverse (FF controller) | success rate: 100% -> 90% | Hurt |
| increase controller size from 200 to 600 units (ablation) | GRU Q-learning · Copying | success rate: 90% -> 30% | Hurt |
| increase controller size from 200 to 600 units (ablation) | GRU Q-learning · 2-row Addition | success rate: 70% -> 10% | Hurt |
| use feed-forward controller instead of GRU (ablation) | Enhanced Q-learning · Reverse | success rate: 0% -> 100% | Helped |
| test on length 1000 instead of 100 | Enhanced Q-learning · Single Digit Multiplication | success rate: 100% -> 0% | Hurt |
| use continuous reward instead of 0/1 reward (ablation) | Q-learning · Walk (FF controller) | median training time (min): 11.1 -> 250.0 | Hurt |
| use continuous reward instead of 0/1 reward (ablation) | Q-learning · Reverse (FF controller) | median training time (min): 3.1 -> 59.7 | Hurt |
Closest papers by meaning
- Learning to Add, Multiply, and Execute Algorithmic Instructions Exactly with Neural Networks (NeurIPS 2025 · similarity 0.83)
- Slower Generalization, Faster Memorization: A Sweet Spot in Algorithmic Learning (NeurIPS 2026 · similarity 0.82)
- Sample Complexity of Algorithm Selection Using Neural Networks and Its Applications to Branch-and-Cut (NeurIPS 2024 · similarity 0.82)
- Algorithmic Capabilities of Random Transformers (NeurIPS 2024 · similarity 0.82)
- Understanding Addition in Transformers (ICLR 2024 · similarity 0.81)
- Towards Learning High-Precision Least Squares Algorithms with Sequence Models (ICLR 2025 · similarity 0.81)
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.