Kristofer Montazeri

Master Thesis
Supervisors: Dr. Shuhang Gu, Dr. Radu Timofte

Data-Free Quantization of Deep Models

Model compression has become a crucial tool for moving deep neural networks (DNNs) onto mobile devices due to their limited storage capacity. We investigate compression of DNNs and focus the data-free quantization of typical architectures. Our findings demonstrate the feasibility of data-free compression and suggest that further research in the area of compression of DNNs is necessary.