Researchers introduce CoCoEvolve for charts, tables and code

A new arXiv paper introduces CoCoEvolve, a self-supervised approach for cross-representation learning across chart images, tabular data and visualization code. The method defines explicit one-to-one correspondences and optimizes models using agreement between representations without additional annotations. The authors also present CoCoEvolve@Train, CoCoEvolve@Test and CoCoEvolve@Eval, an evaluation suite covering all six cross-representation tasks, and report improvements across four benchmarks.