
141Machine Learning on Geospatial Big Data
7.2.1 approaChes to big data feature learning
The most prominent feature learning techniques are those of restricted Boltzmann
machines and autoencoders. Both techniques are articial neural network techniques
that use an unsupervised learning paradigm to perform feature learning. The input
layers are presented with training examples, and an appropriate weight update rule
is applied. The eventual purpose is to arrive at a set of new features that have been
learned and that correspond to some intrinsic properties of the original data. The
computational time and space complexity of both techniques is linea ...