Struct CrossDomainTransfer 

pub struct CrossDomainTransfer<T: Float + Debug + Send + Sync + 'static> { /* private fields */ }

Engine for cross-domain knowledge transfer.

Maintains a registry of domain-specific knowledge and a global shared representation. Supports computing pairwise domain similarities and performing transfer of knowledge from one domain to another.

Implementations

impl<T: Float + Debug + Send + Sync + 'static + ScalarOperand> CrossDomainTransfer<T>

pub fn new(shared_dim: usize) -> Self

Create a new engine with a shared representation of the given dimension.

pub fn transfer_history(&self) -> &[TransferResult<T>]

Get a reference to the transfer history.

pub fn shared_representation(&self) -> &SharedRepresentation<T>

Get a reference to the shared representation.

pub fn register_domain(&mut self, knowledge: DomainKnowledge<T>) -> Result<()>

Register a new domain or update an existing one.

Errors

Returns OptimError::InvalidConfig if the domain’s shared representation dimension does not match the engine’s shared dimension.

pub fn get_registered_domains(&self) -> Vec<&str>

Return the names of all registered domains (sorted for determinism).

pub fn compute_domain_similarity(&self, source: &str, target: &str) -> Result<T>

Compute the cosine similarity between two domains’ shared representations.

The result is in [-1, 1]. If both vectors are zero the similarity is defined as zero.

Errors

Returns OptimError::InvalidState if either domain is not registered.

pub fn transfer( &mut self, source: &str, target: &str, ) -> Result<TransferResult<T>>

Transfer knowledge from source domain to target domain.

The adapted parameters are computed as:

transferred = target_specific
            + transferability * (source_shared - target_shared)

where transferability is (cosine_similarity + 1) / 2 (mapped to [0, 1]).

Errors

Returns OptimError::InvalidState if either domain is not registered, or OptimError::ComputationError if domain-specific parameter dimensions differ.

pub fn update_shared_representation( &mut self, domain_name: &str, gradients: &Array1<T>, lr: T, ) -> Result<()>

Update the shared representation using gradients from a specific domain.

Applies a simple gradient-descent step: shared_features -= lr * gradients and also updates the domain’s own shared-representation snapshot.

Errors

Returns OptimError::InvalidState if the domain is not registered, or OptimError::ComputationError on dimension mismatch.

pub fn get_transferability_matrix(&self) -> Result<Array2<T>>

Compute an NxN pairwise transferability (cosine-similarity) matrix.

Rows and columns are ordered by sorted domain name. The diagonal is 1.0 (self-similarity).

Errors

Returns OptimError::InsufficientData if fewer than two domains are registered.

Trait Implementations

impl<T: Debug + Float + Debug + Send + Sync + 'static> Debug for CrossDomainTransfer<T>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.