Linearity-based neural network compression

Silas Dobler, Florian Lemmerich

In neural network compression, most current methods reduce unnecessary parameters by measuring importance and redundancy. To augment already highly optimized existing solutions, we…