The SqueezeNet model

Description

The SqueezeNet model was introduced by Iandola et al. (https://arxiv.org/pdf/1602.07360.pdf), to reduce the model size and the number of parameters.

The network was made smaller by replacing 3 x 3 filters with 1 x 1 filters as shown here:  

The number of inputs of the 3 x 3 filters has also reduced downsampling of the layers when happening at the higher level, providing large activation maps:

The Core Concepts of Model-Based Design

Model based design in Simulink