Research map / Architectures and efficiency / Attention and transformers
Number Sequence Prediction Problems for Evaluating Computational Powers of Neural Networks
Hyoungwook Nam, Segwang Kim, Kyomin Jung
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 bidirectional instead of unidirectional LSTM in the attention model (ablation) | Attention model · Reverse-order (n=16) | test error rate: 42.0% -> 0.0% | Improved |
| Use bidirectional instead of unidirectional LSTM in the attention model (ablation) | Attention model · Fibonacci | test error rate: 69.3% -> 60.9% | Improved |
| Add attention to LSTM | Digit-level sequence model · Geometric | test error rate: 79.4% -> 62.8% | Improved |
| Add attention to LSTM | Digit-level sequence model · Arithmetic | test error rate: 77.1% -> 77.0% | No clear effect |
| Add stack memory | Digit-level RNN · Reverse-order (n=16) | test error rate: 28.4% -> 0.0% | Improved |
| Add external memory with NTM | Digit-level RNN · Reverse-order (n=16) | test error rate: 28.4% -> 0.0% | Improved |
| Add stack or external memory | Digit-level RNN · Arithmetic | test error rate: 77.1% -> 63.8%; 77.1% -> 65.7% | Improved |
| Add stack or external memory | Digit-level RNN · Fibonacci | test error rate: 80.5% -> 69.4%; 80.5% -> 68.1% | Improved |
| Increase CNN depth for number-level sequences (ablation) | Number-level CNN · Number-level recurrence tasks | prediction error rate | Improved |
| Increase CNN depth for a mixture of primary sequence rules (ablation) | Number-level CNN · Mixed primary sequences | prediction error rate | No clear effect |
| Use deeper models for compound sequence rules (ablation) | Number-level CNN · Compound-rule sequences | prediction error rate | Improved |
| Use ReLU rather than gate activation (ablation) | Number-level CNN · Number-level sequences | accuracy | No clear effect |
Closest papers by meaning
- Limits of Deep Learning: Sequence Modeling through the Lens of Complexity Theory (ICLR 2025 · similarity 0.86)
- Auto-Regressive Next-Token Predictors are Universal Learners (ICML 2024 · similarity 0.86)
- Measuring In-Context Computation Complexity via Hidden State Prediction (ICML 2025 · similarity 0.85)
- Benchmarking Large Language Models with Integer Sequence Generation Tasks (NeurIPS 2025 · similarity 0.85)
- (How) Can Transformers Predict Pseudo-Random Numbers? (ICML 2025 · similarity 0.83)
- Arithmetic Without Algorithms: Language Models Solve Math with a Bag of Heuristics (ICLR 2025 · similarity 0.83)
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.
