arXiv cs.AI· Yangze Liu, Xiao-Long Yin, Zhongyi Han·· 6 天前AI 评分22
研究：训练样本的价值取决于学习器，ResNet-18 与 ViT 的易样本-覆盖样本边界不同
Useful to Whom? The Easy-versus-Coverage Boundary Moves with the Learner
AI 导读
一项选择实验发现，训练样本的价值并非样本固有属性，而是在与学习器的交互中产生。在低分辨率 ImageNet-100 上，ResNet-18 宽度翻倍使易样本与覆盖样本的交叉点从每类 57 个样本移至 85 个；输入从 32 像素增至 64 像素则使其从 85 回到 57，单次 stride 改动即可复现或消除该偏移。在 Vision Transformer 上，覆盖样本在所有测量预算下均占优。
正文 · 原文
Abstract:Is the same data equally useful to every learner? The question is whether the value of a training sample belongs to the sample or arises in its interaction with the learner. We test it with a selection experiment. Under a per-class budget, two kinds of samples compete for the slots: easy samples, which a short training run learns first, and coverage samples, which spread over the class in feature space. Easy samples win at small budgets and coverage samples at large ones, and the budget at which the winner flips, the crossover m0, can be measured. We ask whether m0 is set by the data or by the learner. To separate the two, we freeze the selected subsets and change only the learner that trains on them. On low-resolution ImageNet-100, doubling the width of a ResNet-18 moves the crossover from 57 to 85 samples per class on the same subsets. Enlarging the input grid from 32 to 64 pixels at fixed image information moves it from 85 to 57, and a single stride change reproduces or removes this shift. The grid shift shrinks as width grows. Under the native ImageNet-1k protocol, a 4x range of width moves the boundary by at most a few samples per class. On a Vision Transformer, coverage wins at every measured budget, even with easy subsets that work on the convolutional learner. The budget at which one kind of sample stops being the more useful kind therefore belongs to the learner, and the preferred regime of a selection strategy has to be read against the learner that will use the data.
| Comments: | 22 pages, 9 figures |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.00221 [cs.LG] |
| (or arXiv:2610.00221v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.00221 arXiv-issued DOI via DataCite |
Submission history
From: Yangze Liu [view email]
[v1]
Tue, 22 Sep 2026 05:17:34 UTC (185 KB)
[v2]
Fri, 9 Oct 2026 05:21:11 UTC (2,810 KB)
来源：arXiv cs.AI · arxiv.org