Research map / Large language models / Agents, code and math
Analysing Mathematical Reasoning Abilities of Neural Models
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 |
|---|---|---|---|
| Increase the attentional LSTM's fixed thinking steps from 0 to 16 (ablation) | Attentional LSTM · Mathematics dataset, average across modules | accuracy | Helped |
| Use more than one memory slot (ablation) | Relational recurrent neural network · Mathematics dataset | accuracy | No clear effect |
| Replace fixed thinking steps with Adaptive Computation Time (ablation) | Attentional LSTM · Mathematics dataset | accuracy | Hurt |
| Use a simple LSTM instead of an attentional LSTM (ablation) | LSTM sequence-to-sequence models · Mathematics dataset | accuracy | No clear effect |
| Use a Transformer rather than the best recurrent model | Sequence-to-sequence models · Mathematics dataset, nearly all modules | accuracy | Helped |
| Use a Transformer rather than recurrent models for polynomial manipulation | Sequence-to-sequence models · Polynomial manipulation | accuracy | Helped |
| Train on several terms and extrapolate to inputs with more terms | Neural models · Add/subtract multiple numbers extrapolation | accuracy | Hurt |
| Use the trained Transformer on public British secondary-school mathematics exam questions | Transformer · High-school mathematics questions (40 questions) | correct answers: 14/40 | Mixed |
Closest papers by meaning
- Arithmetic Without Algorithms: Language Models Solve Math with a Bag of Heuristics (ICLR 2025 · similarity 0.91)
- Atomic Thinking of LLMs: Decoupling and Exploring Mathematical Reasoning Abilities (NeurIPS 2025 · similarity 0.89)
- CogMath: Assessing LLMs' Authentic Mathematical Ability from a Human Cognitive Perspective (ICML 2025 · similarity 0.89)
- Evaluating Large Vision-and-Language Models on Children's Mathematical Olympiads (NeurIPS 2024 · similarity 0.88)
- Physics of Language Models: Part 2.1, Grade-School Math and the Hidden Reasoning Process (ICLR 2025 · similarity 0.88)
- Neuro-Symbolic Data Generation for Math Reasoning (NeurIPS 2024 · similarity 0.88)
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.