Consistency-Driven Co-Evolution for Self-Supervised Cross-Representation Learning

2026-08-05

AI Summary

This work addresses the challenge of cross-representational understanding among charts, tables, and visualization code, which is hindered by one-to-many mappings, scarce annotations, and the absence of a unified optimization signal. To this end, we propose CoCoEvolve, a novel framework that introduces, for the first time, a consistency-driven co-evolution mechanism. By establishing one-to-one correspondences among chart–table–code triplets, CoCoEvolve enables joint optimization during both training and inference without requiring additional annotations. The model integrates self-supervised learning, cycle consistency, cross-modal alignment, and test-time optimization within an end-to-end trainable architecture. Extensive experiments across four benchmark datasets and six diverse tasks demonstrate that CoCoEvolve significantly enhances both training efficiency and inference performance, validating its effectiveness and strong generalization capability.

Abstract

As chart images, tabular data, and visualization code play increasingly important roles across diverse domains, cross-representation understanding across these modalities poses fundamental challenges for AI systems: the relationships across representations are inherently \textit{one-to-many}, supervision is ambiguous and costly, and model optimization lacks a principled signal that is both direction-adaptive and representation-generalizable beyond task-specific objectives. We introduce CoCoEvolve to improve consistency across chart, table, and code representations. Instead of treating cross-representation mapping as a one-to-many problem, we define explicit one-to-one correspondences and optimize models using agreement between representations, without additional annotations. During training, CoCoEvolve@Train performs co-evolution across the chart-table-code cycle, while CoCoEvolve@Test applies the same consistency objective at inference time for test-time co-optimization. We also present CoCoEvolve@Eval, an evaluation suite covering all six cross-representation tasks. Across four benchmarks, CoCoEvolve improves performance in both training-time and test-time settings. Our project page: https://xhguo7.github.io/CoCoEvolve/.

Problem

• cross-representation learning
• consistency
• self-supervised learning
• chart-table-code alignment
• one-to-many mapping

Innovation

• consistency-driven
• co-evolution
• self-supervised learning
• cross-representation learning
• test-time optimization