The classification performance of deep neural networks has begun to asymptote at near-perfect levels. However, their ability to generalize outside the training set and their robustness to adversarial attacks have not. In this paper, we make progress on this problem by training with full label distributions that reflect human perceptual uncertainty. We first present a new benchmark dataset which we call CIFAR10H, containing a full distribution of human labels for each image of the CIFAR10 test set. We then show that, while contemporary classifiers fail to exhibit human-like uncertainty on their own, explicit training on our dataset closes this gap, supports improved generalization to increasingly out-of-training-distribution test datasets, and confers robustness to adversarial attacks.

Human categorization is one of the most important and successful targets of cognitive modeling, with decades of model development and assessment using simple, low-dimensional artificial stimuli. However, it remains unclear how these findings relate to categorization in more natural settings, involving complex, high-dimensional stimuli. Here, we take a step towards addressing this question by modeling human categorization over a large behavioral dataset, comprising more than 500,000 judgments over 10,000 natural images from ten object categories. We apply a range of machine learning methods to generate candidate representations for these images, and show that combining rich image representations with flexible cognitive models captures human decisions best. We also find that in the high-dimensional representational spaces these methods generate, simple prototype models can perform comparably to the more complex memory-based exemplar models dominant in laboratory settings.

No abstract

No abstract

No abstract