Negative learning: indirect learning method for noisy data classification with CNN

Abstract

PL), which is a fast and accurate method if the labels are assigned correctly to all images. Accurately labeling a large number of images is daunting and time-consuming, occasionally yielding mismatched labeling. When these noisy labels are present, training with PL will provide wrong information, thus severely degrading performance. Robust training with noisy data is of great practical importance because, when collecting data via online for custom dataset, often there exists data that corresponds to mismatched labels. When robust training with noisy data becomes possible, trimming and evaluating the data is unnecessary, achieving efficiency in time and human resources. To address this issue, we propose an indirect learning method called Negative Learning (NL), in which the CNNs are trained using a complementary label as in "input image does not belong to this complementary label." Because the chances of selecting a true label as a complementary label are low, NL decreases the risk of providing incorrect information. In this dissertation, we demonstrate the distinct advantage of NL for noisy data classification and further propose development complementing NL method.

Convolutional Neural Networks (CNNs) provides excellent performance when used for image classification. The classical method of training CNNs is by labeling images in a supervised manner as in "input image belongs to this label" (Positive Learning