Contrastive Identification of Covariate Shift in Image Data

One possible approach may be to visualize training and test distributions side-by-side (i.e., juxtaposition) using Identifying covariate shift is crucial for making machine learning dimensionality reduction methods (e.g., t-SNE) [1] and show each systems robust in the real world and for detecting training data biases data point as an image thumbnail [4, 26]. However, the scale of that are not reflected in test data. However, detecting covariate shift modern image datasets makes it difficult because we cannot easily is challenging, especially when the data consists of high-dimensional show many images on the projected space [4,8]. Instead of visualizing images, and when multiple types of localized covariate shift affect the distributions of the entire training and test datasets globally, we different subspaces of the data. Although automated techniques aim to intelligently show only local regions of the space, where the can be used to detect the existence of covariate shift, our goal is to locality is informed by the detection algorithm. For example, given help human users characterize the extent of covariate shift in large a test set image highly ranked by a shift detection algorithm (i.e., image datasets with interfaces that seamlessly integrate information deviated from training set distribution), a visualization may show obtained from the detection algorithms. In this paper, we design that many of its similar test images (i.e., local neighborhood) share a and evaluate a new visual interface that facilitates the comparison characteristic (e.g., many faces with sunglasses) while the similar of the local distributions of training and test data. We conduct a training images do not (e.g., no faces with sunglasses).