Microsoft.VisualBasic.DeepLearning 6.0.9762.11310.

Namespaces 17 Types 98 Members 908

01 Namespaces

Microsoft.VisualBasic.MachineLearning

The deep learning namespace root of the sciBASIC# machine learning framework.

Type	Summary	Members
Extensions	Numerically stable activation and normalization helpers shared by the deep learning models.	3

Microsoft.VisualBasic.MachineLearning.CNN

The convolutional neural network: the network model together with its data containers, layers, loss layers and trainers.

Type	Summary	Members
CNNLayerArguments	The "layer specification": a deferred description of a layer that has not been created yet.	4
CNNLayers	Factory functions for building the layer specifications of a CNN network.	17
ConvolutionalNN	A network class holding the layers and some helper functions for training and validation.	17
DataLink	Base class for layer links that hold the input and output DataBlock of a layer.	1
Dimension	The layer dimension data	10
LayerBuilder	Builder that assembles the ordered layer list of a ConvolutionalNN.	23
LayerTypes	Identifies the kind of a CNN layer; the DescriptionAttribute value is the tag persisted in the model file.	38
ReadModelCNN	Deserializes a ConvolutionalNN from the CNN binary model format produced by SaveModelCNN.	1
SaveModelCNN	Serializes a ConvolutionalNN into the CNN binary model format that is read back by ReadModelCNN.	1
Trainer		2
Util		1

Microsoft.VisualBasic.MachineLearning.CNN.data

Data containers for the convolutional neural network.

Type	Summary	Members
BackPropResult	When we have done a back propagation of the network we will receive a result of weight adjustments required to learn.	8
DataBlock	Holding all the data handled by the network.	47
OutputDefinition	This class will hold the definitions that bridge two layers.	7
TrainResult	Created by danielp on 1/27/17.	3

Microsoft.VisualBasic.MachineLearning.CNN.layers

Convolutional neural network layer implementations.

Type	Summary	Members
Conv2DTransposeLayer	Transposed (de)convolution layer: it upsamples a feature map back to a larger spatial size by scattering each input value over the receptive field of every filter.	7
ConvolutionLayer	This layer uses different filters to find attributes of the data that affects the result.	10
DropoutLayer	This layer will remove some random activations in order to defeat over-fitting.	7
FourierFeatureLayer	Based on the paper "Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains" (2020) presented at NeurIPS (https://bmild.github.io/fourfeat/index.…	5
FullyConnectedLayer	Neurons in a fully connected layer have full connections to all activations in the previous layer, as seen in regular Neural Networks.	9
GaussianLayer	Gaussian activation layer.	6
InputLayer	The input layer is a simple layer that will pass the data though and create a window into the full training data set.	9
Layer	A convolution neural network is built of layers that the data traverses back and forth in order to predict what the network sees in the data.	4
LeakyReluLayer	Leaky rectified linear unit activation: f(x) = x for x > 0 and f(x) = leakySlope * x otherwise.	6
LocalResponseNormalizationLayer	This layer is useful when we are dealing with ReLU neurons.	7
MaxoutLayer	Implements Maxout nonlinearity that computes x to max(x) where x is a vector of size group_size.	7
PoolingLayer	This layer will reduce the dataset by creating a smaller zoomed out version.	8
RectifiedLinearUnitsLayer	This is a layer of neurons that applies the non-saturating activation function f(x)=max(0,x).	7
SigmoidLayer	Implements Sigmoid nonlinearity elementwise x to 1/(1+e^(-x)) so the output is between 0 and 1.	6
TanhLayer	Implements Tanh nonlinearity elementwise x to tanh(x) so the output is between -1 and 1.	6

Microsoft.VisualBasic.MachineLearning.CNN.losslayers

Loss layers of the convolutional neural network.

Type	Summary	Members
LossLayer	Base class of the loss layers that terminate a network: it flattens the incoming activations into a vector and computes the loss together with the gradient consumed by the backwar…	13
RegressionLayer	Regression layer is used when your output is an area of data.	7
SoftMaxLayer	This layer will squash the result of the activations in the fully connected layer and give you a value of 0 to 1 for all output activations.	7
SVMLayer	This layer uses the input area trying to find a line to separate the correct activation from the incorrect ones.	7

Microsoft.VisualBasic.MachineLearning.CNN.trainers

Trainers for the convolutional neural network.

Type	Summary	Members
AdaDeltaTrainer	Adaptive delta will look at the differences between the expected result and the current result to train the network.	4
AdaGradTrainer	The adaptive gradient trainer will over time sum up the square of the gradient and use it to change the weights.	3
AdamTrainer	Adaptive Moment Estimation is an update to RMSProp optimizer.	4
NesterovTrainer	Another extension of gradient descent is due to Yurii Nesterov from 1983,[7] and has been subsequently generalized @author Daniel Persson (mailto.woden@gmail.com)	2
SGDTrainer	Stochastic gradient descent (often shortened in SGD), also known as incremental gradient descent, is a stochastic approximation of the gradient descent optimization method for min…	2
TrainerAlgorithm	Trainers take the generated output of activations and gradients in order to modify the weights in the network to make a better prediction the next time the network runs with a da…	16
WindowGradTrainer	This is AdaGrad but with a moving window weighted average so the gradient is not accumulated over the entire history of the run.	2

Microsoft.VisualBasic.MachineLearning.Convolutional

feed-forward phase of deep Convolutional Neural Networks > https://github.com/atasoyhus/CeNiN

Type	Summary	Members
CeNiN	CeNiN (means "fetus" in Turkish) is a minimal implementation of feed-forward phase of deep Convolutional Neural Networks	19
Convolution	Convolutional layer of a CeNiN network: it slides a bank of filters over the padded input and produces one output feature map per filter.	10
ImageProcessor	Provides the bitmap resizing strategies used to fit an arbitrary input image into the fixed spatial size expected by the network's input layer.	0
Input	the layer for image inputs	11
Layer	Base class of all layers in a CeNiN feed-forward network.	24
Output	Terminal layer of a CeNiN network.	8
Pool	Max pooling layer of a CeNiN network: each output value is the maximum activation inside a pool x pool window of the input feature map.	8
ReLU	Rectified linear unit layer: applies f(x) = max(0, x) element wise.	3
SaveModel	Serializes a CeNiN model back into the CeNiN binary file format, the reverse of loading a model with LoadFile.	1
SoftMax	Softmax output layer that converts the incoming activations into a probability distribution.	3
Solver	High level inference helpers that run a loaded CeNiN network over an image and return the ranked class predictions.	1

Microsoft.VisualBasic.MachineLearning.Convolutional.ImageProcessor

Type	Summary	Members
ResizingMethod	Specifies how a source bitmap is mapped onto the network input size.	4

Microsoft.VisualBasic.MachineLearning.NeuralNetwork

Feed forward neural network models.

Type	Summary	Members
HiddenLayersView	Read-only view over all hidden layers of a network.	5
Network	Object model for artificial neural network computation.	28
NetworkKernel	Uses a fully connected ConvolutionalNN as the single compute kernel of the NeuralNetwork namespace.	2
NetworkLayerView	Read-only view of a single neural network layer (input, hidden or output).	4
Netz	Feed forward neural network used for regression analysis.	25

Microsoft.VisualBasic.MachineLearning.RNN

Recurrent neural networks: the recurrent layers, cells and training helpers.

Type	Summary	Members
Alphabet	Immutable set of symbols mapped indices.	7
BasicRNN	RNN that uses integer indices as inputs and outputs.	1
CharacterSampleable	Network that can be sampled for a sequence of characters.	2
CharLevelRNN	RNN that can use both indices, and characters as inputs/outputs.	4
CharRNN	Command line helpers for training, sampling and snapshotting a character level recurrent network.	5
IntegerSampleable	Network that can be sampled for a sequence of integer indices.	2
Math	Math helper functions	7
Matrix	A dense M x N matrix of doubles used by the character level RNN implementation.	40
MultiLayerCharLevelRNN	Multi layer character level RNN: a character level network backed by a stack of RNN layers.	13
MultiLayerRNN	Multi layer RNN: a stack of RNNLayer instances where the output of one layer feeds the next.	10
Options	Application options.	12
Random	Helper functions for randomness.	4
RNN	A recurrent neural network.	5
RNNLayer	An RNN Layer with support for multi-layer networks.	20
RNNTrainer		12
SingleLayerCharLevelRNN	Single layer character level RNN.	11
SingleLayerRNN	Single layer RNN.	11
StringTrainingSet	Immutable training set for a character level RNN: a block of text together with the alphabet extracted from it.	7
Trainable	Trainable neural network.	1
TrainingSet	Training set for sequences.	3
Utils	Utility functions for working with the jagged Double()() arrays used by the RNN matrix type.	3

Microsoft.VisualBasic.MachineLearning.Transformer

The transformer architecture: attention layers, positional encodings and the transformer model.

Type	Summary	Members
DecoderLayer	One decoder layer: masked self attention, cross attention and a position wise feed forward network, each wrapped in a residual connection and layer normalization.	9
DecoderStack	A stack of DecoderLayer instances, applied one after another.	7
Embedding	Use a learned embedding layer to reduce the size of the word embedding space.	17
EncoderLayer	One encoder layer: multi head self attention followed by a position wise feed forward network, each wrapped in a residual connection and layer normalization.	6
EncoderStack	A stack of EncoderLayer instances, applied one after another.	6
FeedForwardNetwork	Position wise feed forward network: two fully connected layers with a ReLU activation in between.	6
MultiHeadAttention	Scaled dot product multi head attention, implemented with the hand written operators of TensorOps and an explicit backward pass.	7
Optimizer	实现 Adam 优化器（参数 + 同形梯度累加器）。	7
OutputLayer	Produce a flat array with the same dimension as the number of words in the dictionary	8
TensorOps	Transformer 迁移辅助算子集合：补齐 Tensor 相对旧 AD 张量缺失的 N 维算子，并为每个算子提供显式反向传播实现。	30
TextProcessing	Text preprocessing helpers used by the translation demo: loading parallel sentence pairs, tokenizing sentences, computing sequence lengths and adding the start / stop markers.	8
TransformerModel	Transformer architecture as described in "Attention is all you need"	7

Microsoft.VisualBasic.MachineLearning.Transformer.DecoderLayer

Type	Summary	Members
Cache	Forward intermediates of one decode step, required by the backward pass.	38

Microsoft.VisualBasic.MachineLearning.Transformer.EncoderLayer

Type	Summary	Members
Cache	Forward intermediates of one encode pass, required by the backward pass.	26

Microsoft.VisualBasic.MachineLearning.Transformer.FeedForwardNetwork

Type	Summary	Members
Cache	前向传播的中间量缓存，供反向传播使用。	6

Microsoft.VisualBasic.MachineLearning.Transformer.MultiHeadAttention

Type	Summary	Members
Cache	Forward intermediates of the attention computation, required by the backward pass.	18

Microsoft.VisualBasic.MachineLearning.Transformer.OutputLayer

Type	Summary	Members
Cache	Forward intermediates of the output projection, required by the backward pass.	6

Microsoft.VisualBasic.MachineLearning.Transformer.TransformerModel

Type	Summary	Members
DecoderStep	一个解码步的前向记录，供 BPTT 反向使用。	6