Combating Label Noise in Image Data Using MultiNET Flexible Confident Learning

Deep neural networks (DNNs) have been used successfully for many image classification problems. One of the most important factors that determines final efficiency a DNN is correct construction training set. Erroneously labeled images can degrade accuracy and additionally lead to unpredictable model behavior, reducing reliability. In this paper, we propose MultiNET, novel method automatic detection noisy labels within datasets. MultiNET an adaptation current state-of-the-art confident learning method. contrast original, our aggregates outputs multiple DNNs allows adjustment sensitivity. We conduct exhaustive evaluation, incorporating four widely datasets (CIFAR10, CIFAR100, MNIST, GTSRB), eight architectures, variety noise scenarios. Our results demonstrate significantly outperforms