Problem

The use of a dataset comprised of a multi-dimensional feature space for neural network training and subsequent prediction not only requires excessive processing and memory resources, but also takes a long time.

Solution

The input feature space is converted into a smaller feature space by learning a compressed representation of the input feature space.

Application

The input dataset is compressed into a smaller dataset through the use of an AutoEncoder neural network.

 

 

 

 

 

 

 

 

 

 

 

A data scientist prepares a dataset comprised of a large number of features (1). The dataset is then used to train a neural network (2, 3). The resulting network takes longer to train and carry out predictions, and requires increased computing resources and memory (4).