Selecting Samples on Graphs: A Unified Dataset Pruning Framework for Lossless Training Acceleration
arXiv:2606.12913v2 Announce Type: replace Abstract: The rapid growth of modern training datasets has significantly increased computational cost, motivating dataset pruning~(DP) methods which retain only a subset of informative samples to reduce training cost. Existing pruning criteria typically rely on either intrinsic signals that assess samples independently or extrinsic signals that promote diversity via pairwise relations. While effective in their own specific regimes, each captures only one aspect of sample utility and lacks robustness across different pruning ratios or data distribution.
The continuous growth in dataset size for AI training necessitates more efficient methods to manage computational costs without compromising model performance.
Achieving 'lossless training acceleration' through smarter data pruning directly addresses the escalating energy and compute demands of large-scale AI, impacting profitability and sustainability.
This framework offers a unified approach to dataset pruning, moving beyond fragmented methods that struggle with diverse data or varying pruning ratios, indicating more robust and adaptable AI training efficiency.
- · AI compute providers
- · Hyperscalers
- · AI model developers
- · Data scientists
- · Inefficient AI training practices
- · Undifferentiated compute services
Reduced computational resource usage for training large AI models.
Faster iteration and deployment cycles for AI applications, leading to quicker market entry.
Democratization of advanced AI development by lowering the barrier of entry for compute-constrained entities.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG