Research map / Architectures and efficiency / RNNs and normalization
Visual Learning of Arithmetic Operation
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 |
|---|---|---|---|
| Train end-to-end from pictures of Arabic numerals to picture outputs | Feed-forward network · Addition | incorrect digit rate: 1.9% | Helped |
| Train the same visual end-to-end approach for subtraction | Feed-forward network · Subtraction | incorrect digit rate: 3.2% | Helped |
| Apply the visual end-to-end approach to multiplication | Feed-forward network · Multiplication | incorrect digit rate: 71.5% | Hurt |
| Use five hidden layers for visual multiplication | Feed-forward network · Multiplication | incorrect digit rate: 71.5% | Hurt |
| Apply visual end-to-end training to Roman numeral addition | Feed-forward network · Roman numeral addition | incorrect digit rate: 74.3 % | Hurt |
| Train on strongly Gaussian-noised input and output pictures | Visual addition network · Noisy addition | OCR digit error rate: 9.8% OCR; 3.2% human inspection | Mixed |
| Replace picture inputs and outputs with 1-hot vectors (ablation) | Addition network · Addition | incorrect digit rate: 1.9% -> 1.7% | No clear effect |
| Replace picture inputs and outputs with 1-hot vectors (ablation) | Subtraction network · Subtraction | incorrect digit rate: 3.2% -> 2.1% | No clear effect |
| Replace picture inputs and outputs with 1-hot vectors (ablation) | Multiplication network · Multiplication | incorrect digit rate: 71.5% -> 37.6% | Helped |
| Replace picture inputs and outputs with 1-hot vectors (ablation) | Roman numeral addition network · Roman numeral addition | incorrect digit rate: 74.3 % -> 0.7 % | Helped |
Closest papers by meaning
- Arithmetic Without Algorithms: Language Models Solve Math with a Bag of Heuristics (ICLR 2025 · similarity 0.82)
- Understanding Addition in Transformers (ICLR 2024 · similarity 0.81)
- DARR: A Dual-Branch Arithmetic Regression Reasoning Framework for Solving Machine Number Reasoning (AAAI 2025 · similarity 0.81)
- Teaching Arithmetic to Small Transformers (ICLR 2024 · similarity 0.81)
- Interpreting and Improving Large Language Models in Arithmetic Calculation (ICML 2024 · similarity 0.80)
- Transformers Can Do Arithmetic with the Right Embeddings (NeurIPS 2024 · similarity 0.80)
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.