ailayer_conv2d Struct Reference

General Conv2D layer structure. More...

#include <ailayer_conv2d.h>

Data Fields
ailayer_t	base
	Inherited field members from general ailayer struct.
uint16_t	result_shape [4]
	The shape of the result.
Layer configuration
Required configuration parameters for the layer These fields have to be configured by the user before calling the initializer function.
uint32_t	filter_count
	Number of filters ( \( C_{out} \)).
uint16_t	kernel_size [2]
	Size of the kernels ( \( [H_{kernel}, W_{kernel}] \)).
uint16_t	stride [2]
	The stride in the direction of height and width ( \( [S_{h}, S_{w}] \)).
uint16_t	dilation [2]
	The dilation in the direction of height and width ( \( [D_{h}, D_{w}] \)).
uint16_t	padding [2]
	The (symmetric) zero padding in the direction of height and width ( \( [P_{h}, P_{w}] \)).
int8_t	channel_axis
	Index of the channel axis (1 for channels first and -1 or 3 for channels last).
Trainable parameters
Data fields for the trainable parameters (weights, bias) of the layer
aitensor_t	weights
	Tensor containing the weights.
aitensor_t	bias
	Tensor containing the bias weights.
uint16_t	weights_shape [4]
	Weights tensor shape.
uint16_t	bias_shape [1]
	Bias weights tensor shape.
aitensor_t *	trainable_params [2]
	Pointer to the weights and bias (which are the trainable parameters).
aitensor_t *	gradients [2]
	Gradients structure for the backpropagation algorithm.
void *	optimem [2]
	Memory field used by the optimizer for training.
Math functions
Required data type specific math functions
void(*	conv2d_fwd )(const aitensor_t *input, const uint16_t stride[2], const uint16_t dilation[2], const uint16_t padding[2], const aitensor_t *weights, const aitensor_t *bias, int8_t channel_axis, void *work_space, aitensor_t *output)
	Required math function: 2D-Convolution. More...
void(*	conv2d_bwd )(const aitensor_t *input, const uint16_t stride[2], const uint16_t dilation[2], const uint16_t padding[2], const aitensor_t *output, int8_t channel_axis, void *work_space, aitensor_t *d_weights)
	Required math function: Gradients of 2D-Convolution with respect to the weights. More...
void(*	conv2d_bwd_full )(const aitensor_t *delta_out, const uint16_t stride[2], const uint16_t dilation[2], const uint16_t padding[2], const aitensor_t *weights, int8_t channel_axis, void *work_space, aitensor_t *delta_in)
	Required math function: Gradients of 2D-Convolution with respect to the input. More...
void(*	tensor_add )(const aitensor_t *a, const aitensor_t *b, aitensor_t *result)
	Required math function: Element wise tensor addition. More...
void(*	sum_channelwise )(const aitensor_t *x, int8_t channel_axis, aitensor_t *result)
	Required math function: Channel-wise sum. More...

Detailed Description

General Conv2D layer structure.

Field Documentation

conv2d_bwd

void(* conv2d_bwd) (const aitensor_t *input, const uint16_t stride[2], const uint16_t dilation[2], const uint16_t padding[2], const aitensor_t *output, int8_t channel_axis, void *work_space, aitensor_t *d_weights)

Required math function: Gradients of 2D-Convolution with respect to the weights.

Requires a math function that calculates the gradients with respect to the weights \( \partial w = \mathrm{d} L / \mathrm{d} w \).

\[ \partial w = x_{in} \ast delta_{out} \]

Parameters

input	Input data with dimension \( [N,C_{in},H_{in},W_{in}] \) (channels first) or \( [N,H_{in},W_{in},C_{in}] \) (channels last)
stride	The stride in the direction of height and width
dilation	The dilation in the direction of height and width
padding	The (symmetric) zero padding in the direction of height and width
output	Gradients backpropagated from the following layer with dimension \( [N,C_{out},H_{out},W_{out}] \) (channels first) or \( [N,H_{out},W_{out},C_{out}] \) (channels last)
channel_axis	Index of the channel axis (1 for channels first and -1 or 3 for channels last).
work_space	Pointer to a work space buffer for intermediate results.
d_weights	Output gradients of the weights with dimension \( [C_{out},C_{in},H_{kernel},W_{kernel}] \) (channels first) or \( [C_{out},H_{kernel},W_{kernel},C_{in}] \) (channels last)

conv2d_bwd_full

void(* conv2d_bwd_full) (const aitensor_t *delta_out, const uint16_t stride[2], const uint16_t dilation[2], const uint16_t padding[2], const aitensor_t *weights, int8_t channel_axis, void *work_space, aitensor_t *delta_in)

Required math function: Gradients of 2D-Convolution with respect to the input.

Requires a math function that calculates the gradients with respect to the input \( delta_{in} = \mathrm{d} L / \mathrm{d} x_{in} \).

\[ delta_{in} = delta_{out} \ast' w \]

\( \cdot \ast' \cdot \) is a transposed convolution.

Parameters

delta_out	Gradients backpropagated from the following layer with dimension \( [N,C_{in},H_{in},W_{in}] \) (channels first) or \( [N,H_{in},W_{in},C_{in}] \) (channels last)
stride	The stride in the direction of height and width
dilation	The dilation in the direction of height and width
padding	The (symmetric) zero padding in the direction of height and width
weights	Convolution kernels with dimension \( [C_{out},C_{in},H_{kernel},W_{kernel}] \) (channels first) or \( [C_{out},H_{kernel},W_{kernel},C_{in}] \) (channels last)
channel_axis	Index of the channel axis (1 for channels first and -1 or 3 for channels last).
work_space	Pointer to a work space buffer for intermediate results.
delta_in	Resulting input gradients for backpropagation to the previous layer with dimension \( [N,C_{in},H_{in},W_{in}] \) (channels first) or \( [N,H_{in},W_{in},C_{in}] \) (channels last)

conv2d_fwd

void(* conv2d_fwd) (const aitensor_t *input, const uint16_t stride[2], const uint16_t dilation[2], const uint16_t padding[2], const aitensor_t *weights, const aitensor_t *bias, int8_t channel_axis, void *work_space, aitensor_t *output)

Required math function: 2D-Convolution.

Requires a math function that performs a 2D-convolution and adds a channel-wise bias:

\[ x_{out} = x_{in} \ast w + b \]

Parameters

input	Input data with dimension \( [N,C_{in},H_{in},W_{in}] \) (channels first) or \( [N,H_{in},W_{in},C_{in}] \) (channels last)
stride	The stride in the direction of height and width
dilation	The dilation in the direction of height and width
padding	The (symmetric) zero padding in the direction of height and width
weights	Convolution kernels with dimension \( [C_{out},C_{in},H_{kernel},W_{kernel}] \) (channels first) or \( [C_{out},H_{kernel},W_{kernel},C_{in}] \) (channels last)
bias	Bias with dimension \( C_{out} \)
channel_axis	Index of the channel axis (1 for channels first and -1 or 3 for channels last).
work_space	Pointer to a work space buffer for intermediate results.
output	Output after convolution with dimension \( [N,C_{out},H_{out},W_{out}] \) (channels first) or \( [N,H_{out},W_{out},C_{out}] \) (channels last)

sum_channelwise

void(* sum_channelwise) (const aitensor_t *x, int8_t channel_axis, aitensor_t *result)

Required math function: Channel-wise sum.

Requires a math function that calculates the sum of all elements of each channel c. The result tensor is 1D.:

\[ result_c = \sum_i(x_{ci}) \]

tensor_add

void(* tensor_add) (const aitensor_t *a, const aitensor_t *b, aitensor_t *result)

Required math function: Element wise tensor addition.

Requires a math function that adds two tensors element wise:

\[ result = a + b \]

---

The documentation for this struct was generated from the following file:

• ailayer_conv2d.h