A preview is not available for this record, please engage by choosing from the available options ‘download’ or ‘view’ to engage with the material
Description
Brijraj Singh This work optimizes an already compressed pretrained model by using a residual connection to reduce the number of floating point operations. It locates blocks in a pretrained architecture that have a relatively lesser knowledge quotient and then bypasses those blocks through a Shunt connection. This connection combines an encoder and decoder and is relatively more optimized in terms of the number of parameters.