The cookie-related information is fully under our control. These cookies are not used for any purpose other than those described here. Unibo policy
In this work, we carry out an analysis of the neural network called the autoencoder, in the case of simple geometric shapes. In particular, we describe the precise mechanisms which allow the autoencoder to encode and decode a simple geometric shapes, the disk. This is done using an ablation study, where we describing the exact solution of the decoding process in the case of a network without biases. We show that the learning process approximates this solution. We also carry out an experimental investigation of the generalisation capacity of this autoencoder and discuss the best regularisation approaches to improve this generalisation capacity.
This is poster number 38 in Poster Session