nuget
api documents
Neurons in a fully connected layer have full connections to all activations in the previous layer, as seen in regular Neural Networks. Their activations can hence be computed with a matrix multiplication followed by a bias offset.
@author Daniel Persson (mailto.woden@gmail.com)
01 Syntax
02 Methods
| Name | Overloads | Summary |
|---|---|---|
| .ctor | 2 | Creates an empty layer, used by the deserializer. |
| forward | 1 | Runs the affine transformation y = Wx + b of this layer. |
| PackWeights | 1 | 把各个输出神经元的权值按行拼成一个 (out_depth, num_inputs) 的矩阵。 |
| backward | 1 | Computes the gradients with respect to the weights, the biases and the input of this layer. |
| ToString | 1 | Returns a short description of this layer. |
03 Properties
| Name | Overloads | Summary |
|---|---|---|
| BackPropagationResult | 1 | Gets the weight and bias parameter blocks of this layer. |
| Type | 1 | Gets the kind of this layer, always LayerTypes.FullyConnected. |
04 Fields
| Name | Overloads | Summary |
|---|---|---|
| weightsPacked | 1 | 最近一次前向传播时按后端布局打包好的权值矩阵, 形状 (out_depth, num_inputs)。 |
05 Members
Creates an empty layer, used by the deserializer.
Int32)Creates a fully connected layer with the given number of neurons.
| Name | Type | Description |
|---|---|---|
def | OutputDefinition | The shared output definition that carries the flattened input size. |
num_neurons | Int32 | Number of neurons, i.e. the number of outputs of this layer. |
Boolean)Runs the affine transformation y = Wx + b of this layer.
| Name | Type | Description |
|---|---|---|
db | DataBlock | The input data block, interpreted as a column vector. |
training | Boolean | Ignored; the fully connected layer behaves the same in both modes. |
The output activations.
把各个输出神经元的权值按行拼成一个 (out_depth, num_inputs) 的矩阵。
每个 FullyConnectedLayer.filters(i) 内部是连续的 num_inputs 个权值, 目标矩阵的每一行也是连续的, 所以这里可以按行整块拷贝, 不需要逐元素重排。
Computes the gradients with respect to the weights, the biases and the input of this layer.
Returns a short description of this layer.
A text of the form full_connected(N).
Gets the weight and bias parameter blocks of this layer.
Gets the kind of this layer, always LayerTypes.FullyConnected.
最近一次前向传播时按后端布局打包好的权值矩阵, 形状 (out_depth, num_inputs)。
本层的权值原本是按"每个输出神经元一个 DataBlock"分开存放的, 与后端 MatMul 要求的 矩阵形状不匹配; 这里在每次前向时按行拼成一个矩阵, 反向传播也需要它(的转置), 所以一并缓存下来。因为权值在训练过程中一直被就地更新, 每次前向都重新打包, 不存在过期问题。