How Does the Combined Risk Affect the Performance of Unsupervised Domain Adaptation Approaches?

Unsupervised domain adaptation (UDA) aims to train a target classifier with labeled samples from the source and unlabeled domain. Classical UDA learning bounds show that risk is upper bounded by three terms: risk, distribution discrepancy, combined risk. Based on assumption small fixed value, methods based this bound only minimizing estimators of discrepancy. However, may increase when both estimators, which makes uncontrollable. Hence cannot achieve ideal performance if we fail control To key challenge takes root in unavailability address challenge, propose method named E-MixNet. E-MixNet employs enhanced mixup, generic vicinal distribution, pseudo-labeled calculate proxy Experiments can effectively curb Furthermore, added into loss functions four representative methods, their also improved.