The paper introduces Minmax-CF, a label-aware characteristic-function selector for distilling single-cell datasets into traceable real-cell coresets. It formulates compression as a discrete min-max problem, uses entropy-regularized maximization to emphasize poorly preserved directions, and greedily selects cells and genes under separate budgets. Across five coarse-lineage benchmarks and five compression budgets, it reportedly retains 95.3% of the full-reference macro-F1 on average. It also preserves source-cell indices and original gene symbols, and outperforms size-matched synthetic PCA-Centroid and Distribution Matching baselines in 24 of 25 comparisons against each.
No heat snapshots are available in the last 24 hours.