In this topic, we are working with machine learning applications to represent large volumes of data from computer simulations. The data involved can be scalar (temperature, pressure, etc.), vector (velocity) or tensor fields. The basic idea is to use neural networks to generate a feature vector that can represent the patterns (phenomena) of interest in a space with a much smaller dimension than the original data space. Such a representation, in addition to reducing the computational cost of subsequent operations, can also be more robust in relation to insignificant variations in the data. In addition, techniques for video processing based on machine learning are being researched to analyze the temporal evolution of patterns with a view to applications in fluid analysis and aerial image processing.