Integrating Feedforward Design into a Generative Network to Synthesize Supplementary Training Data for Object Classification

Training a neural network typically requires large amounts of data. Yet, gathering such amounts of data may pose a problem. This can be addressed by augmenting a data set, i. e., by adding artificial data to the original data set. One method of augmentation is generating additional images, for example with a generative adversarial network. However, this technique still requires a lot of data. This work tackles this problem by implementing the so-called feedforward design as an alternative neural network training method which supposedly requires fewer training data than regular backpropagation. We apply feedforward design to the task of image generation. The results are evaluated by augmenting original data sets to form a new training set for a neural network classifier.