DPCTGANGenerator<T>

Models & Types · AiDotNet.NeuralNetworks.SyntheticData

Differentially Private CTGAN (DP-CTGAN) for generating synthetic tabular data with formal (epsilon, delta)-differential privacy guarantees.

Inheritance: NeuralNetworkBase<T> → NeuralSyntheticTabularGeneratorBase<T> → DPCTGANGenerator<T>

Implements: ISyntheticTabularGenerator<T>

For Beginners

DP-CTGAN works exactly like CTGAN but with a “privacy shield”:

**Standard CTGAN training step:**DP-CTGAN training step:

If you provide custom layers in the architecture, those will be used directly for the generator network. If not, the network creates industry-standard CTGAN layers.

Example usage:

How It Works

This implementation follows the standard neural network architecture pattern with:

  • Proper inheritance from NeuralNetworkBase
  • Layer-based architecture using ILayer components
  • Engine-based tensor operations for CPU/GPU acceleration
  • Full autodiff and JIT compilation support

DP-CTGAN modifies the CTGAN training loop to provide differential privacy:

  • Per-sample gradient clipping: Each sample’s gradient is clipped to a fixed L2 norm
  • Gaussian noise: Calibrated noise added to aggregated clipped gradients
  • Privacy accounting: Tracks cumulative privacy cost using moments accountant
  • Early stopping: Training halts when privacy budget is exhausted

Constructors

Constructor Summary
DPCTGANGenerator() Initializes a new DP-CTGAN generator with the specified architecture.
DPCTGANGenerator(NeuralNetworkArchitecture<T> architecture, DPCTGANOptions<T> options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction)

Properties

Property Summary
DPCTGANOptions<T> Options { get; } Gets the DP-CTGAN-specific options.
IReadOnlyList<ColumnMetadata> Columns { get; }
bool IsFitted { get; set; }
double CumulativeEpsilon { get; } Gets the cumulative privacy cost (epsilon) spent so far during training.

Methods

Method Summary
Dictionary<string, T> GetFeatureImportance()
Matrix<T> Generate(int numSamples, Vector<T> conditionColumn, Vector<T> conditionValue)
ModelMetadata<T> GetModelMetadata()
Task FitAsync(Matrix<T> data, IReadOnlyList<ColumnMetadata> columns, int? epochs, CancellationToken ct)
void Fit(Matrix<T> data, IReadOnlyList<ColumnMetadata> columns, int? epochs) Fits the DP-CTGAN generator to the provided real tabular data with differential privacy.