CoCoEvolve: Self-Supervised Cross-Representation Learning for Charts, Tables, and Code

A new AI framework called CoCoEvolve aims to improve cross-representation understanding across chart images, tabular data, and visualization code. The approach, detailed in a paper on arXiv (2608.04926), addresses the challenge of one-to-many relationships between these modalities by defining explicit one-to-one correspondences and optimizing models using agreement between representations without additional annotations. The method introduces two variants: CoCoEvolve@Train, which performs co-evolution across the chart-table-code cycle during training, and CoCoEvolve@Test, which applies the same consistency objective at inference time. This work is significant for AI systems that need to interpret and generate visualizations, as it provides a principled signal that is direction-adaptive and representation-generalizable beyond task-specific objectives. The paper was announced as a cross-type submission on arXiv, indicating it spans multiple research areas.