Supplementary Material -- Towards Reliable Model Selection for Unsupervised Domain Adaptation: An Empirical Study and A Certified Baseline Mi Luo 3

We first prove the first inequality using Jensen's inequality, which states that for a real-valued, convex function φ with its domain as a subset of R and numbers t Next, we leverage the property of inequalities to prove the second inequality. Substituting the NLL loss with any strongly convex loss function would still uphold the proposition. We introduce the respective computation involved in the existing model selection approaches. However, this method has limited effectiveness in scenarios with severe domain shifts between the source and target domains. Additionally, it introduces additional hyperparameters for dataset splitting, which can further complicate the model selection process.