Binary neural network

Binary neural network is an artificial neural network, where commonly used floating-point weights are replaced with binary ones.^[1] It largely saving the storage and computation, serves as a technique for deploying deep models on resource-limited devices. Usage of binary values can bring up to 58 times speedup,^[2] while accuracy and information capacity of binary neural network can be manually controlled.^[3] Binary neural networks do not achieve the same accuracy as their full-precision counterparts, but improvements are being made to close this gap.

References

↑ Courbariaux, M.; Bengio, Y.; David, J.-P. (2015). "BinaryConnect: training deep neural networks with binary weights during propagation". NIPS. arXiv:1511.00363.
↑ Rastegari, M.; Ordonez, V.; Redmon, J.; Farhadi, A. (2016). "XNOR-Net: ImageNet classification using binary convolutional neural networks". ECCV. arXiv:1603.05279.
↑ Ignatov, D.; Ignatov, A. (2020). "Controlling information capacity of binary neural network". Pattern Recognition Letters. arXiv:2008.01438. doi:10.1016/j.patrec.2020.07.033.

[1] Courbariaux, M.; Bengio, Y.; David, J.-P. (2015). "BinaryConnect: training deep neural networks with binary weights during propagation". NIPS. arXiv:1511.00363.

[2] Rastegari, M.; Ordonez, V.; Redmon, J.; Farhadi, A. (2016). "XNOR-Net: ImageNet classification using binary convolutional neural networks". ECCV. arXiv:1603.05279.

[3] Ignatov, D.; Ignatov, A. (2020). "Controlling information capacity of binary neural network". Pattern Recognition Letters. arXiv:2008.01438. doi:10.1016/j.patrec.2020.07.033.

[1]

[2]

[3]