Are all training examples equally valuable?
Agata Lapedriza Massachusetts Institute of Technology Universitat Oberta de Catalunya Computer Vision Center Hamed Pirsiavash Massachusetts Institute of Technology Zoya Bylinskii Massachusetts Institute of Technology Antonio Torralba Massachusetts Institute of Technology
Abstract
When learning a new concept, not all training examples may prove equally useful for training: some may have higher or lower training value than others. The goal of this paper is to bring to the attention of the vision community the following considerations: (1) some examples are better than others for training detectors or classifiers, and (2) in the presence of better examples, some examples may negatively impact performance and removing them may be beneficial. In this paper, we propose an approach for measuring the training value of an example, and use it for ranking and greedily sorting examples. We test our methods on different vision tasks, models, datasets and classifiers. Our experiments show that the performance of current state-of-the-art detectors and classifiers can be improved when training on a subset, rather than the whole training set.
中文速览
训练一个视觉识别或检测模型时，人们默认把所有标注数据都喂进去，但实际上并非每张训练图都同等有用——有些样本甚至会拖累模型表现。论文提出了一种衡量"训练价值（training value）"的方法：把每个正样本单独训练成一个小分类器，看它在整个训练集上的检测精度，以此给每个样本打分排序；在此基础上，作者还设计了一种贪心前向筛选算法，逐步挑出最能提升模型性能的样本子集。在 PASCAL VOC 2007 目标检测和 SUN 场景分类等多个任务上，结合 HOG、GIST、视觉词袋等特征以及线性SVM、核SVM、LDA等分类器的实验均表明，用筛选后的子集训练，性能普遍高于使用完整训练集。这一发现提示计算机视觉社区：数据质量的筛选与排序同样重要，盲目堆砌训练数据不仅无益，有时反而有害。
原文 arXiv:1311.6510;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1311.6510v1