Training Compact Neural Networks with Binary Weights and Low Precision Activations

Explore this paper's citation graph

Summary

The proposed network with binary weights and low-bitwidth activations, named Group-Net, outperform the previous best binary neural network in terms of accuracy as well as saving more than 5 times computational complexity on ImageNet with ResNet-18 and Res net-50.

Type
preprint
Published
2018-08-08
Cited by
14
References
44
Access
Open access

Keywords

Computer science, Feature (linguistics), Binary number, Set (abstract data type), Artificial neural network

References

Cited by

Related papers