Calculate deep convolution neurAl network on Cell Unit

This is a deep learning framework for cnn which using xnor method to accelerate calculating convolution layer on cpu. More details CACUE.

Features:

• BIT cell calculating units,fast cpu running.
• using bit in order to accelerate the calculating performance on cpu.You'll find this version is ~X10 fast on cpu.
• GPU mode supported for trainning large-scaled model.
• caffe level precision in 32bits.
• nicely portable header only
• only dependency needed just include boost(dynamic_bitset) if you need bit method.
• just include mycnn.h and write your model in C++. There is nothing to install.
• squeezed model size ,achieve a ~X32 reduction in convolution layer.
• cross platform supported
• running on both linux and windows.

Layers support:

for this version

• layer:

• average_pooling_layer

• batch_normalization_layer

• convolution_layer

• eltwise_layer

• inner_product_layer

• max_pooling_layer

• relu_layer

• sigmoid_layer

• softmax_with_loss_layer

• bit layer:

• bin_activation_layer

• bit_convolution_layer

Bit blob

We use dynamic_bitset which supplied by boost to binary the 32 bits parameters, Bin_blob is created for the binaried data flow in this framework.Each layer can be created on two kinds of blobs,flexiable for more bit logical calculating.

Model design

Build a cnn network like what you did in Caffe.You may easily create a CNN mode in CACU if you are a Caffe user.

References

[1] A Krizhevsky, I Sutskever, GE Hinton. Imagenet classification with deep convolutional neural networks.. Advances in neural information processing systems. 2012: 1097-1105.

[2] Rastegari M, Ordonez V, Redmon J, et al. XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks. arXiv preprint arXiv:1603.05279, 2016.

[3] S Ioffe, C Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift.. arXiv preprint arXiv:1502.03167, 2015.

[4] Courbariaux M, Bengio Y. Binarynet: Training deep neural networks with weights and activations constrained to+ 1 or-1. arXiv preprint arXiv:1602.02830, 2016.

[5] Jia Y, Shelhamer E, Donahue J, et al. Caffe: Convolutional architecture for fast feature embedding arXiv preprint arXiv:1408.5093, 2014.