February 2018
Intermediate to advanced
378 pages
10h 14m
English
Another popular approach to reduce the size of a neural network is via SVD. SVD is applied to pre-adjusted neural networks, and therefore reduces the number of parameters in the network. After reducing the number of parameters, an unconventional backpropagation algorithm is used to train the models restructured by SVD, which has lower time complexity than the conventional BP algorithm. Experimental results have shown almost 2x improved speed with zero loss in accuracy, and around 4x improvement with minor loss in accuracy.
Additional reading: you can explore several other approaches adopted by tech giants for the mobile platform:
Read now
Unlock full access