Deep Learning and Historical Collections
Deep convolutional neural networks have led to breakthrough results in numerous machine learning tasks such as the classification of images in huge data sets, like ImageNet ; they have provided the framework for unsupervised control-policy-learning in the mastering by computers of sample human tasks, like Atari games; and have led to the defeat of the world champion, in the complex and computationally intractable, game of Go, a decade before computer scientists thought it possible. All of these applications first perform feature extraction on large data sets and then feed the results into a trainable classifier based on deep convolutional neural networks. This paper presents an introduction to a framework for the building of a feature extractor that employs large convolutional neural networks to identify and extract layer features from large sets of digitized historical maps that could be used in environmental, urban planning and development studies.
