Generalizing Numerical Reasoning in Table Data through Operation Sketches and Self-Supervised Learning

📰 ArXiv cs.AI

Learn to improve numerical reasoning in table data using operation sketches and self-supervised learning for better generalization and robustness

advanced Published 25 Apr 2026

Action Steps

Apply header anonymization to reduce lexical memorization in table data
Create operation sketches to provide structural reasoning guidance
Implement self-supervised learning for continual pre-training of numerical reasoning models
Test TaNOS framework on various datasets to evaluate its effectiveness
Compare the performance of TaNOS with supervised fine-tuning (SFT) methods

Who Needs to Know This

Data scientists and machine learning engineers working on table data analysis can benefit from this approach to improve model robustness and generalization

Key Insight

💡 Operation sketches and self-supervised learning can enhance model generalization and robustness in numerical reasoning tasks

Key Takeaways

Learn to improve numerical reasoning in table data using operation sketches and self-supervised learning for better generalization and robustness

Full Article

Title: Generalizing Numerical Reasoning in Table Data through Operation Sketches and Self-Supervised Learning

Abstract:
arXiv:2604.21495v1 Announce Type: cross Abstract: Numerical reasoning over expert-domain tables often exhibits high in-domain accuracy but limited robustness to domain shift. Models trained with supervised fine-tuning (SFT) on specific datasets tend to rely on header-operation shortcuts rather than structural reasoning. We introduce TaNOS, a continual pre-training framework comprising three components: (i) header anonymization to reduce lexical memorization, (ii) operation sketches that provide

Read full paper → ← Back to Reads