Search papers, labs, and topics across Lattice.
This paper introduces Mapping the Concept Landscape (MCL), a novel framework for data pruning that utilizes explicit sample-level graphs to represent image-caption pairs, capturing fine-grained semantic interactions. By constructing a comprehensive dataset-level graph, MCL quantifies the global distribution and rarity of semantic concepts, enabling a greedy algorithm to maximize coverage of under-represented concepts during the pruning process. Experimental results show that MCL outperforms existing methods in pruning efficiency while offering a transparent and interpretable selection process.
Pruning data with MCL not only boosts efficiency but also reveals hidden semantic structures that traditional methods overlook.
Existing data pruning methods predominantly rely on high-dimensional feature embeddings to measure sample importance. However, these compressed vectors often obscure fine-grained semantic interactions, leading to suboptimal coverage of rare semantic concepts in the pruned subsets. In this paper, we propose Mapping the Concept Landscape (MCL), a novel structural perception framework for transparent data pruning. Instead of abstract embeddings, we represent each image-caption pair as an explicit sample-level graph comprising entities, events, and attributes. By integrating these individual graphs into a comprehensive dataset-level graph, we characterize the global distribution of semantic concepts and quantify their rarity across the entire corpus. Based on this structured perception, we develop a greedy concept-coverage maximization algorithm that iteratively selects samples to maximize the marginal gain of high-value, under-represented concepts. Experimental results on various benchmarks demonstrate that our method not only achieves superior pruning efficiency compared to state-of-the-art methods but also provides a transparent and interpretable audit trail for the selection process.
