Deep Learning

const size_t features = 64;
const float keep_probability = 0.8;

// Generative network.
auto g_i = Input({6, 6, 3});
auto g_1 = Dropout(g_i, keep_probability);
auto g_2 = BatchNormalization(g_1);

auto __3 = Deconvolution2D(g_2, {5, 5}, features, 2);
auto __4 = LeakyRelu(__3);
auto __5 = Dropout(__4, keep_probability);
auto __6 = BatchNormalization(__5);

// Layer 1
auto g_3 = Deconvolution2D(__6, {5, 5}, features, 1);
auto g_4 = LeakyRelu(g_3);
auto g_5 = Dropout(g_4, keep_probability);
auto g_6 = BatchNormalization(g_5);

// Layer 2
auto g_7 = Deconvolution2D(g_6, {5, 5}, features, 1);
auto g_8 = LeakyRelu(g_7);
auto g_9 = Dropout(g_8, keep_probability);
auto g_10 = BatchNormalization(g_9);

// Layer 3
auto g_11 = Deconvolution2D(g_10, {5, 5}, 1, 1);
auto g_o = Sigmoid(g_11);

// Build a descriminative network.
auto d_i = Input({27, 27});

// Layer 1
auto d_1 = Convolution2D(d_i, {5, 5}, features, 2);
auto d_2 = LeakyRelu(d_1);
auto d_3 = Dropout(d_2, keep_probability);

// Layer 2
auto d_4 = Convolution2D(d_3, {5, 5}, features, 1);
auto d_5 = LeakyRelu(d_4);
auto d_6 = Dropout(d_5, keep_probability);

// Layer 3
auto d_7 = Convolution2D(d_6, {5, 5}, features, 1);
auto d_8 = LeakyRelu(d_7);
auto d_9 = Dropout(d_8, keep_probability);

// Layer 3
auto d_10 = Linear(d_9, {128});
auto d_11 = LeakyRelu(d_10);

// Layer 4
auto d_12 = Linear(d_11, {2});
auto d_o = Softmax(d_12)