mlsanityData to 14 Oct 2026

Math Neurosurgery: Isolating Language Models' Math Reasoning Abilities Using Only Forward Passes

arXiv 2410.16930 ↗First posted 22 Oct 2024cs.CL cs.AIcited by 10

On the map: LLM reasoning → Mathematical reasoning

Abstract (arXiv)

Math reasoning is an active area of Large Language Model (LLM) research because it is a hallmark of artificial intelligence and has implications in several domains, including math education. However, few works have explored how math reasoning is encoded within LLM parameters and if it is a skill that can be isolated within models. Doing so could allow targeted intervention to improve math performance without altering non-math behavior and foster understanding of how models encode math reasoning. We introduce Math Neurosurgery (MathNeuro), a computationally efficient method we use to isolate math-specific parameters in LLMs using only forward passes. MathNeuro builds on existing work by using weights and activations to calculate parameter importance, but isolates math-specific parameters by filtering out those important for general language tasks. Through pruning parameters MathNeuro identifies, we delete a LLM's math reasoning ability without significantly impacting its general language ability. Scaling the identified parameters by a small constant improves a pretrained or instruction-tuned LLM's performance by 4-17% on GSM8K and 5-35% on MATH while leaving non-math behavior unaltered. MathNeuro is also data efficient: most of its effectiveness holds when identifying math-specific parameters using a single sample. MathNeuro highlights the potential for future work to intervene on math-specific parameters.

Key facts. Cited by 10 papers in the map (data through 14 Oct 2026). Placed in Mathematical reasoning, one of 17 directions in LLM reasoning. Builds on 34 of its 51 references that are in the map. First posted on arXiv 22 Oct 2024.

Builds on

34 of 51 references.

Closest ancestors · 4

Same direction, newest first

  1. 2024An Investigation of Neuron Activation as a Unified Lens to Explain Chain-of-Thought Eliciting Arithmetic Reasoning of LLMs18 Jun
  2. 2024Common 7B Language Models Already Possess Strong Math Capabilities7 Mar
  3. 2024GSM-Plus: A Comprehensive Benchmark for Evaluating the Robustness of LLMs as Mathematical Problem Solvers29 Feb
  4. 2024Large Language Models for Mathematical Reasoning: Progresses and Challenges31 Jan

Other references · 30

  1. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  2. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  3. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  4. 2019HellaSwag: Can a Machine Really Finish Your Sentence?cited by 4,100
  5. 2024Gemma 2: Improving Open Language Models at a Practical Sizecited by 1,998
  6. 2024Large Language Monkeys: Scaling Inference Compute with Repeated Samplingcited by 991
  7. 2023A Simple and Effective Pruning Approach for Large Language Modelscited by 800
  8. 2019MathQA: Towards Interpretable Math Word Problem Solving with Operation-Based Formalismscited by 662
  9. 2023An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-tuningcited by 639
  10. 2023Textbooks Are All You Need II: phi-1.5 technical reportcited by 565
  11. 2023Finding Neurons in a Haystack: Case Studies with Sparse Probingcited by 334
  12. 2024DeepSeek-Prover: Advancing Theorem Proving in LLMs through Large-Scale Synthetic Datacited by 234
18 more
  1. 2024Many-Shot In-Context Learningcited by 206
  2. 2024How do Large Language Models Handle Multilingualism?cited by 150
  3. 2017Learning to Generate Reviews and Discovering Sentimentcited by 148
  4. 2024Language-Specific Neurons: The Key to Multilingual Capabilities in Large Language Modelscited by 126
  5. 2018What Is One Grain of Sand in the Desert? Analyzing Individual Neurons in Deep NLP Modelscited by 124
  6. 2020MuTual: A Dataset for Multi-Turn Dialogue Reasoningcited by 115
  7. 2018Identifying and Controlling Important Neurons in Neural Machine Translationcited by 111
  8. 2024On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neuronscited by 83
  9. 2024LLM2LLM: Boosting LLMs with Novel Iterative Data Enhancementcited by 70
  10. 2024Lean Copilot: Large Language Models as Copilots for Theorem Proving in Leancited by 58
  11. 2024LoFiT: Localized Fine-tuning on LLM Representationscited by 57
  12. 2024Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertisecited by 50
  13. 2024Interpreting and Improving Large Language Models in Arithmetic Calculationcited by 45
  14. 2024Whispering Experts: Neural Interventions for Toxicity Mitigation in Language Modelscited by 32
  15. 2024Towards Understanding Multi-Task Learning (Generalization) of LLMs via Detecting and Exploring Task-Specific Neuronscited by 25
  16. 2023Unified Low-Resource Sequence Labeling by Sample-Aware Dynamic Sparse Finetuningcited by 21
  17. 2024Does Knowledge Localization Hold True? Surprising Differences Between Entity and Relation Perspectives in Language Modelscited by 9
  18. 2020Analyzing Redundancy in Pretrained Transformer Modelscited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 6; 2026: 4.

10 papers.

  1. 2025Feature Extraction and Steering for Enhanced Chain-of-Thought Reasoning in Language Modelscited by 13
  2. 2025Leveraging the true depth of LLMscited by 7
  3. 2025Task-Circuit Quantization: Leveraging Knowledge Localization and Interpretability for Compressioncited by 7
  4. 2025L2V-CoT: Cross-Modal Transfer of Chain-of-Thought Reasoning via Latent Interventioncited by 5
  5. 2025Fine, I'll Merge It Myself: A Multi-Fidelity Framework for Automated Model Mergingcited by 4
  6. 2025StreetMath: Study of LLMs' Approximation Behaviorscited by 2
  7. 2026CreativityNeuro: Steering Language Model Weights to Improve Divergent Thinking and Reduce Mode Collapsecited by 2
  8. 2026LLM Parameters for Math Across Languages: Shared or Separate?cited by 1
  9. 2026When Tokenization is Secretly Output Supervisioncited by 0
  10. 2026Neuron-Guided Fine-Tuning: Unlocking Efficient Alignment Mechanisms for Large Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024From Next-Token to Mathematics: The Learning Dynamics of Mathematical Reasoning in Language Models
  2. 2024Neuro-Symbolic Data Generation for Math Reasoning
  3. 2025Benchmarking Large Language Models for Diagnosing Students' Cognitive Skills from Handwritten Math Work
  4. 2025Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning
  5. 2024Arithmetic Without Algorithms: Language Models Solve Math With a Bag of Heuristics
  6. 2025Improving Rule-based Reasoning in LLMs using Neurosymbolic Representations
  7. 2024Physics of Language Models: Part 2.1, Grade-School Math and the Hidden Reasoning Process
  8. 2025Evaluating Mathematical Reasoning Across Large Language Models: A Fine-Grained Approach