Class CertifiedDefenseOptions<T>

Namespace

AiDotNet.Models.Options

Assembly

AiDotNet.dll

Configuration options for certified defense mechanisms.

public class CertifiedDefenseOptions<T>

Type Parameters

T

The numeric data type used for calculations (e.g., float, double).

Inheritance

object

CertifiedDefenseOptions<T>

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

These options control certified robustness methods that provide provable guarantees about model predictions under adversarial perturbations.

For Beginners: These settings control how the "guaranteed protection" works. You can adjust how many samples to use, how tight the guarantees should be, and what certification method to apply.

Properties

BatchSize

Gets or sets the batch size for certification.

public int BatchSize { get; set; }

Property Value

int

The batch size, defaulting to 100.

Remarks

For Beginners: Processing multiple inputs at once speeds up certification. Larger batches are faster but use more memory.

CertificationMethod

Gets or sets the certification method to use.

public string CertificationMethod { get; set; }

Property Value

string

The method name, defaulting to "RandomizedSmoothing".

Remarks

For Beginners: Different certification methods have different trade-offs: - RandomizedSmoothing: Works for any model, scalable - IBP (Interval Bound Propagation): Faster but might be looser - CROWN: Tight bounds but more computationally expensive

ConfidenceLevel

Gets or sets the confidence level for certification.

public double ConfidenceLevel { get; set; }

Property Value

double

The confidence level (0-1), defaulting to 0.99.

Remarks

For Beginners: This is how confident you want to be in the guarantee. 0.99 means 99% confidence that the guarantee holds. Higher confidence requires more samples.

NoiseSigma

Gets or sets the noise standard deviation for randomized smoothing.

public double NoiseSigma { get; set; }

Property Value

double

The noise sigma, defaulting to 0.25.

Remarks

For Beginners: This controls how much random noise is added. Larger sigma gives stronger robustness guarantees but might reduce accuracy.

NormType

Gets or sets the norm type for certification.

public string NormType { get; set; }

Property Value

string

The norm type, defaulting to "L2".

Remarks

For Beginners: This defines what "small perturbation" means for certification. L2 norm is common for randomized smoothing.

NumSamples

Gets or sets the number of samples for randomized smoothing.

public int NumSamples { get; set; }

Property Value

int

The number of samples, defaulting to 1000.

Remarks

For Beginners: Randomized smoothing makes predictions by averaging over many noisy versions of the input. More samples give tighter guarantees but take longer.

RandomSeed

Gets or sets the random seed for reproducible certification.

public int? RandomSeed { get; set; }

Property Value

int?

The random seed, or null for non-deterministic random generation.

Remarks

When set to a specific value, the certification process produces reproducible results, which is useful for testing and debugging. When null, each certification run uses different random samples for proper statistical validity.

For Beginners: If you need reproducible results (e.g., for testing), set this to a specific number. For actual robustness certification, leave it as null so each run uses different random samples.

UseTightBounds

Gets or sets whether to use tight bounds computation.

public bool UseTightBounds { get; set; }

Property Value

bool

True for tight bounds, false for faster looser bounds (default: false).

Remarks

For Beginners: Tight bounds give stronger guarantees but take longer to compute. Looser bounds are faster but might certify fewer examples.