On generalization in moment-based domain adaptation
This work provides theoretical guarantees for domain adaptation, which is important for improving model performance in target domains with limited data, though it is incremental as it builds on existing discrepancy measures.
The paper tackles the problem of deriving generalization bounds for domain adaptation algorithms under practical conditions, resulting in bounds based on finitely many moments and smoothness.
Domain adaptation algorithms are designed to minimize the misclassification risk of a discriminative model for a target domain with little training data by adapting a model from a source domain with a large amount of training data. Standard approaches measure the adaptation discrepancy based on distance measures between the empirical probability distributions in the source and target domain. In this setting, we address the problem of deriving generalization bounds under practice-oriented general conditions on the underlying probability distributions. As a result, we obtain generalization bounds for domain adaptation based on finitely many moments and smoothness conditions.