Algorithm 1: Weighted Subdomain Adaptation Network (WSAN)

Model: Feature generator G; Auxiliary classifier CA; Classifier C.

Input: Labeled source data {Xs, Cs} and unlabeled target data Xt.

For i in epochs:

Step 1: The feature generator G outputs the high-dimensional features of the two domains and inputs them into the feature generator G and classifier C.

Step 2: Auxiliary classifier CA obtains the class-level weights. The classifier gives prediction probability output on the target samples and obtains sample-level weights to guide WLMMD to perform subdomain adaptation.

Step 3: Train the feature generator G and classifier C to obtain the optimal parameters ${\widehat{\theta }}_G$ and ${\widehat{\theta }}_C$ by minimizing F(${\theta }_G,{\theta }_C$);

Step 4: Train the auxiliary classifier CA to obtain the optimal parameters ${\widehat{\theta }}_{C_A}$ by minimizing F(${\theta }_{C_A})$;