Core Functions

Model Creation

chain - Create a neural network model
parameters - Access model parameters

Layer Types

Static - Input layer specifying dimensions
Dense - Fully connected layer
Conv - Convolutional layer
MaxPool - Max pooling layer
Flatten - Flatten multi-dimensional data

Forward Pass

preallocate - Create forward pass cache
set_inputs! - Set input data
forward! - Execute forward pass
get_outputs - Extract model outputs

Backward Pass

preallocate_grads - Create gradient cache
backprop! - Execute backward pass
gradients - Access computed gradients

Loss Functions

MSELoss - Mean squared error loss
LogitCrossEntropyLoss - Cross entropy loss for classification

Activation Functions

relu - ReLU activation
sigmoid - Sigmoid activation
tanh_fast - Fast tanh activation

GPU Support

gpu - Move models/data to GPU

Function Index