Learning advisor networks for noisy image classification

In this paper we introduce the novel concept of advisor network to tackle the noisy image classification task. Deep neural networks (DNN) are prone to performance reduction and overfitting problems on noisy training data. Weighting loss methods tend to completely remove the influence of noisy examples during the training. This prevents DNNs from learning wrong associations between images and their correct labels but reduces the amount of data actually used, especially if most of the examples are noisy. Differently our method weighs the feature extracted directly from the main model without altering the loss value of each data. The advisor helps the DNNs to focus only on some part of information present in noisy examples allowing them to leverage that data as well. We trained it with a meta learning strategy so that it can adapt throughout the training of the main model. We tested our method on CIFAR10 and CIFAR100 with synthetic noise, and on Clothing1M that contains real-world noise, reporting state-of-the-art results.