论文阅读 【CVPR-2022】 Active Learning by Feature Mixing

论文阅读 【CVPR-2022】 Active Learning by Feature Mixing

通过特征混合进行主动学习

studyai.com

搜索论文: Active Learning by Feature Mixing

摘要(Abstract)

The promise of active learning (AL) is to reduce labelling costs by selecting the most valuable examples to annotate from a pool of unlabelled data.

主动学习（AL）的愿景是通过从未标注的数据池中选择最有价值的样例进行标注来减少标注的成本。

Identifying these examples is especially challenging with high-dimensional data (e.g. images, videos) and in low-data regimes.

识别这些高价值样例在高维数据（如图像、视频）和低数据状态下尤其具有挑战性。

In this paper, we propose a novel method for batch AL called ALFA-Mix.

在本文中，我们提出了一种新型的批量主动学习方法，称为ALFA-Mix。

We identify unlabelled instances with sufficiently-distinct features by seeking inconsistencies in predictions resulting from interventions on their representations.

我们通过寻找对其表征的干预所导致的预测中的不一致，来识别具有足够独特性特征的未标记的实例。

We construct interpolations between representations of labelled and unlabelled instances then examine the predicted labels.

我们在有标签的实例和无标签的实例的表征之间构建插值，然后检查预测出的标签。

We show that inconsistencies in these predictions help discovering features that the model is unable to recognise in the unlabelled instances.

我们展示了，这些预测中的不一致有助于发现模型在未标记的实例中无法识别的特征。

We derive an efficient implementation based on a closed-form solution to the optimal interpolation causing changes in predictions.

我们根据引起预测变化的最优插值的解析解，得出了一个有效的实现方案。

Our method outperforms all recent AL approaches in 30 different settings on 12 benchmarks of images, videos, and non-visual data.

我们的方法在图像、视频和非视觉数据的12个基准上，在30个不同的设置中胜过了最近所有的AL方法。

The improvements are especially significant in low-data regimes and on self-trained vision transformers, where ALFA-Mix outperforms the stateof-the-art in 59% and 43% of the experiments respectively.

在低数据量的情况下，以及在自训练的视觉变换器上，ALFA-Mix分别在59%和43%的实验中超过了最先进的方法，这种改进尤为显著。