Caffe XNOR-nets

Czech title:Caffe XNOR-nets
Reseach leader:Hradiš Michal
Agency:Fingera s.r.o.
Start:2017-03-01
End:2017-06-30
Keywords:convolutional networks, CNN,  NXOR-nets
Annotation:
Evaluation of NXOR-nets and opensource tools for training of these networks.
NXOR-nets are defined in Rastegari et al.: XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks.
The evaluated propertties of NXOR-nets are the degradation of classification accuracy due to transformation to binary networks and the possibilities of compensating this degradation by enlarging the network.
Further, the project will result in free open source implementations of layers needed to train NXOR-nets in caffe framework using GPU.
Project outputs:
* Technická zpráva o chování NXOR sítí
* OpenSource implementace vrstev nutných pro trénování NXOR sítí s využitím GPU
* Implementace dopředného průchodu těchto vrstev pro PC CPU jako funkce v čistém C++ bez výrazných optimalizací

Your IPv4 address: 54.162.166.214
Switch to IPv6 connection

DNSSEC [dnssec]