Attributes, or semantic features, have gained popularity in the past few\nyears in domains ranging from activity recognition in video to face\nverification. Improving the accuracy of attribute classifiers is an important\nfirst step in any application which uses these attributes. In most works to\ndate, attributes have been considered to be independent. However, we know this\nnot to be the case. Many attributes are very strongly related, such as heavy\nmakeup and wearing lipstick. We propose to take advantage of attribute\nrelationships in three ways: by using a multi-task deep convolutional neural\nnetwork (MCNN) sharing the lowest layers amongst all attributes, sharing the\nhigher layers for related attributes, and by building an auxiliary network on\ntop of the MCNN which utilizes the scores from all attributes to improve the\nfinal classification of each attribute. We demonstrate the effectiveness of our\nmethod by producing results on two challenging publicly available datasets.\n