PickGraphLayer<T>
Layers · AiDotNet.NeuralNetworks.Layers
PICK’s graph learning-convolutional network (reference GLCN): it learns a soft adjacency over the
document’s text segments from their embeddings, then runs numLayersPickGcnLayer
steps seeded by an embedding of the six box-relation features.
Inheritance: LayerBase<T> → PickGraphLayer<T>
Implements: IShapeContract
For Beginners
This decides which text boxes on a page should talk to each other, then lets them exchange information so that, for example, a “Total:” label informs the number beside it.
How It Works
Graph learning (reference GraphLearningLayer):
x_hat = x P, with no bias andlearningDimcolumns.A = softmax_j(LeakyReLU(w . |x_hat_i - x_hat_j|)) + 1e-10.- The graph-learning loss is
sum_ij exp(A_ij + eta ||x_hat_i - x_hat_j||) / N^2 + gamma ||A||_F.
The pairwise distance is computed as sqrt(s_ij + I_ij) - I_ij. That is the same value, and it keeps
the gradient finite on the diagonal, where the reference’s norm of an exact zero vector has an
infinite derivative.
Constructors
| Constructor | Summary |
|---|---|
PickGraphLayer(int dim, int learningDim, int numLayers, double eta, double gamma) |
Creates the network. |
Properties
| Property | Summary |
|---|---|
bool SupportsTraining { get; } |
Methods
| Method | Summary |
|---|---|
Dictionary<string, string> GetParameterRoles() |
Returns parameter roles for per-role optimizer configuration (e.g., weight decay exemption for biases). |
IReadOnlyList<ILayer<T>> GetSubLayers() |
|
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank) |
|
IReadOnlyList<Tensor<T>> GetTrainableParameters() |
Returns all trainable parameter tensors marked with [TrainableParameter]. |
void ResetState() |
|
void SetTrainableParameters(IReadOnlyList<Tensor<T>> parameters) |
Replaces trainable parameter tensors (e.g., with ParameterBuffer views). |
void ZeroGrad() |
Clears all gradient fields discovered by convention ({paramName}Gradient). |
Fields
| Field | Summary |
|---|---|
int RelationFeatures |
Box-relation features per node pair (reference: x_ij, y_ij, w_i/h_i, h_j/h_i, w_j/h_i, len_j/len_i). |