GRASP: Geometry-aware Residual Alignment for Scalable Pretraining Data Attribution
About
Scalable data attribution methods typically assign isolated utility scores to individual training examples. This prevalent additive assumption fundamentally fails to capture critical subset dynamics, including data redundancy and complementary coverage. In this work, we reframe attribution as subset-level counterfactual utility prediction and introduce GRASP, an interaction-aware surrogate. Grounded in a theoretical smoothness lower bound, GRASP explicitly models subset interactions through a quadratic geometric penalty. To achieve pretraining-scale efficiency without relying on hidden oracle tuning, we couple low-dimensional feature sketches with a strictly finite lower-confidence bound selection protocol. Extensive subset-retraining evaluations demonstrate that GRASP decisively outperforms existing scalable baselines. It more than doubles the task-level rank correlation for counterfactual subset fidelity while reducing upfront artifact construction costs by nearly an order of magnitude. Downstream diagnostics further show that this scoring mechanism transfers to language model curation and cross-domain vision selection, establishing a robust foundation for optimizing massive pretraining corpora.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Noisy label detection | CIFAR-10 1k subset 10% label noise | AUC (%)80.32 | 8 | |
| Vision data selection | CIFAR-10 1k 10% label noise | Accuracy (20% Selected Data)56.1 | 8 | |
| Malicious or compromised fine-tuning example detection | Misleading information and safety risks dataset | F1 Score30 | 6 | |
| Related-domain retrieval | CC-10B WebOrganizer Topic labels (candidate pool) | Mean Normalized Recall@10007.121 | 6 | |
| Related-domain retrieval | CC-10B WebOrganizer Format labels (candidate pool) | Mean Normalized Recall@10004.1 | 6 | |
| Counterfactual Subset-Utility Evaluation | BasicSkills Common Knowledge | Spearman Correlation (ρ)0.312 | 4 | |
| Counterfactual Subset-Utility Evaluation | BasicSkills Logical Reasoning | Spearman Correlation ($
ho$)0.457 | 4 | |
| Counterfactual Subset-Utility Evaluation | OpenBookQA | Spearman Correlation0.35 | 4 | |
| Counterfactual Subset-Utility Evaluation | CommonsenseQA | Spearman Correlation ($
ho$)0.26 | 4 | |
| Counterfactual Subset-Utility Evaluation | SciQ | Spearman Correlation (ρ)0.424 | 4 |