Improving Robustness of Tabular Retrieval via Representational Stability

2026-04-27

AI Summary

This work addresses the sensitivity of table retrieval to serialization formats such as CSV and HTML, which causes semantically identical tables to yield substantially different embeddings when represented in distinct formats. Treating each format as a noisy view of a shared underlying semantic structure, the authors propose a centroid-based geometric correction method. This approach constructs a canonical representation from the centroid of embeddings across multiple formats and introduces a lightweight residual bottleneck adapter combined with covariance regularization to align single-format embeddings toward this centroid. The adapter is fine-tuned while keeping the base encoder frozen, enabling compatibility with various models including MPNet, BGE-M3, ReasonIR, and SPLADE. Experiments demonstrate that the centroid representation consistently outperforms any single-format embedding, and the adapter significantly enhances robustness in dense retrieval, though it yields limited gains for sparse retrieval.

Technology Category

• Machine Learning: Representation Learning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Transformer-based table retrieval systems flatten structured tables into token sequences, making retrieval sensitive to the choice of serialization even when table semantics remain unchanged. We show that semantically equivalent serializations, such as $\texttt{csv}$, $\texttt{tsv}$, $\texttt{html}$, $\texttt{markdown}$, and $\texttt{ddl}$, can produce substantially different embeddings and retrieval results across multiple benchmarks and retriever families. To address this instability, we treat serialization embedding as noisy views of a shared semantic signal and use its centroid as a canonical target representation. We show that centroid averaging suppresses format-specific variation and can recover the semantic content common to different serializations when format-induced shifts differ across tables. Empirically, centroid representations outrank individual formats in aggregate pairwise comparisons across $\texttt{MPNet}$, $\texttt{BGE-M3}$, $\texttt{ReasonIR}$, and $\texttt{SPLADE}$. We further introduce a lightweight residual bottleneck adapter on top of a frozen encoder that maps single-serialization embeddings towards centroid targets while preserving variance and enforcing covariance regularization. The adapter improves robustness for several dense retrievers, though gains are model-dependent and weaker for sparse lexical retrieval. These results identify serialization sensitivity as a major source of retrieval variance and show the promise of post hoc geometric correction for serialization-invariant table retrieval. Our code, datasets, and models are available at $\href{https://github.com/KBhandari11/Centroid-Aligned-Table-Retrieval}{https://github.com/KBhandari11/Centroid-Aligned-Table-Retrieval}$.

Problem

• tabular retrieval
• serialization sensitivity
• representational stability
• embedding variance
• format invariance

Innovation

• representational stability
• centroid alignment
• table retrieval
• serialization invariance
• residual bottleneck adapter