TabSieve: Explicit In-Table Evidence Selection for Tabular Prediction

2026-02-12

AI Summary

Existing tabular prediction models struggle to effectively leverage relevant in-table rows as few-shot evidence and are highly susceptible to noise, leading to unstable performance. This work proposes TabSieve, a novel framework that decouples and jointly optimizes evidence selection and prediction for the first time: it explicitly selects a small set of informative table rows as auditable evidence and subsequently makes predictions based on this curated evidence. The approach introduces a TAB-GRPO reinforcement learning strategy coupled with a dynamic task advantage balancing mechanism and is trained on TabSieve-SFT-40K, a synthetically generated high-quality supervised fine-tuning dataset supporting joint classification and regression tasks. Evaluated across 75 classification and 52 regression benchmarks, TabSieve achieves average accuracy gains of 2.92% and regression performance improvements of 4.45%, significantly enhancing both robustness and interpretability.

Technology Category

• Search and Optimization: Learning to Search
• Reasoning under Uncertainty: Relational Probabilistic Models
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

Tabular prediction can benefit from in-table rows as few-shot evidence, yet existing tabular models typically perform instance-wise inference and LLM-based prompting is often brittle. Models do not consistently leverage relevant rows, and noisy context can degrade performance. To address this challenge, we propose TabSieve, a select-then-predict framework that makes evidence usage explicit and auditable. Given a table and a query row, TabSieve first selects a small set of informative rows as evidence and then predicts the missing target conditioned on the selected evidence. To enable this capability, we construct TabSieve-SFT-40K by synthesizing high-quality reasoning trajectories from 331 real tables using a strong teacher model with strict filtering. Furthermore, we introduce TAB-GRPO, a reinforcement learning recipe that jointly optimizes evidence selection and prediction correctness with separate rewards, and stabilizes mixed regression and classification training via dynamic task-advantage balancing. Experiments on a held-out benchmark of 75 classification and 52 regression tables show that TabSieve consistently improves performance across shot budgets, with average gains of 2.92% on classification and 4.45% on regression over the second-best baseline. Further analysis indicates that TabSieve concentrates more attention on the selected evidence, which improves robustness to noisy context.

Problem

• tabular prediction
• in-table evidence
• few-shot learning
• noisy context
• evidence selection

Innovation

• TabSieve
• evidence selection
• tabular prediction
• reinforcement learning
• few-shot reasoning