Layers

Morph provides built-in neural network layers through the NN module.

---

Available Layers

Layer	Description	Creation
NN.Linear(in, out)	Fully-connected layer	NN.Linear(784, 128)
NN.Conv2D(in, out, kernel)	2D convolution	NN.Conv2D(3, 16, 3)
NN.BatchNorm(features)	Batch normalization	NN.BatchNorm(128)
NN.Dropout(rate)	Dropout regularization	NN.Dropout(0.5)

---

Layer Usage

layer is NN.Linear(784, 128);
output is layer.Forward(input);

---

Activation Functions

Function	Usage
NN.ReLU(x)	Rectified Linear Unit
NN.Sigmoid(x)	Sigmoid activation
NN.Tanh(x)	Hyperbolic tangent
NN.Softmax(x)	Softmax normalization

---

Building a Network

MyNetwork class {
    conv1 is NN.Conv2D(1, 32, 3);
    bn1 is NN.BatchNorm(32);
    fc1 is NN.Linear(32 * 26 * 26, 10);
    drop1 is NN.Dropout(0.25);

    Forward method(x as Tensor<float>) as Tensor<float> {
        x is NN.ReLU(bn1.Forward(conv1.Forward(x)));
        x is drop1.Forward(x);
        return NN.Softmax(fc1.Forward(x));
    }
}

---

Next Steps

• Training — Training loop and optimization