Beyond Categorical Label Representations for Image Classification
Beyond Categorical Label Representations for Image Classification
复制标题
DOI：
--
复制
发表时间：
2021-04
期刊：
影响因子：
--
通讯作者：
Boyuan Chen-;Yu Li;Sunand Raghupathi;H. Lipson
中科院分区：
文献类型：
--
作者：
Boyuan Chen-;Yu Li;Sunand Raghupathi;H. Lipson
我们发现，我们选择的表示数据标签的方式会对训练模型的质量产生深远的影响。例如，训练图像分类器来回归音频标签而不是传统的分类概率会产生更可靠的分类。考虑到音频标签比简单的数字概率或文本更复杂，这个结果令人惊讶。我们假设，高维，高熵的标签表示通常更有用，因为它们提供了更强的错误信号。我们用来自各种标签表示的证据支持这一假设，包括常数矩阵、谱图、混洗谱图、高斯混合和各种维度的均匀随机矩阵。我们的实验表明，在标准图像分类任务中，高维、高熵标签的准确性与文本（分类）标签相当，但通过我们的标签表示学习的特征在各种对抗性攻击下表现出更强的鲁棒性，并且在训练数据量有限的情况下表现出更好的有效性。这些结果表明，标签表征可能发挥更重要的作用比以前认为的。该项目的网站位于\url{https：//www.creativemachineslab.com/label-representation.html}。
We find that the way we choose to represent data labels can have a profound effect on the quality of trained models. For example, training an image classifier to regress audio labels rather than traditional categorical probabilities produces a more reliable classification. This result is surprising, considering that audio labels are more complex than simpler numerical probabilities or text. We hypothesize that high dimensional, high entropy label representations are generally more useful because they provide a stronger error signal. We support this hypothesis with evidence from various label representations including constant matrices, spectrograms, shuffled spectrograms, Gaussian mixtures, and uniform random matrices of various dimensionalities. Our experiments reveal that high dimensional, high entropy labels achieve comparable accuracy to text (categorical) labels on the standard image classification task, but features learned through our label representations exhibit more robustness under various adversarial attacks and better effectiveness with a limited amount of training data. These results suggest that label representation may play a more important role than previously thought. The project website is at \url{https://www.creativemachineslab.com/label-representation.html}.