COSDA: Counterfactual-based Susceptibility Risk Framework for Open-Set Domain Adaptation

Open-Set Domain Adaptation (OSDA) aims to transfer knowledge from the labeled source domain to the unlabeled target domain that contains unknown categories, thus facing the challenges of domain shift and unknown category recognition. While recent works have demonstrated the potential of causality for domain alignment, little exploration has been conducted on causal-inspired theoretical frameworks for OSDA. To fill this gap, we introduce the concept of Susceptibility and propose a novel Counterfactual-based susceptibility risk framework for OSDA, termed COSDA. Specifically, COSDA consists of three novel components: (i) a Susceptibility Risk Estimator (SRE) for capturing causal information, along with comprehensive derivations of the computable theoretical upper bound, forming a risk minimization framework under the OSDA paradigm; (ii) a Contrastive Feature Alignment (CFA) module, which is theoretically proven based on mutual information to satisfy the Exogeneity assumption and facilitate cross-domain feature alignment; (iii) a Virtual Multi-unknown-categories Prototype (VMP) pseudo-labeling strategy, providing label information by measuring how similar samples are to known and multiple virtual unknown category prototypes, thereby assisting in open-set recognition and intra-class discriminative feature learning. Extensive experiments demonstrate that our approach achieves state-of-the-art performance.

Machine learning systems often encounter samples with shifting features or previously unseen classes in new environments, leading to significant performance degradation. This challenge has driven the development of neural network adaptation algorithms. From the perspective of causality, such degradation often arises when models rely on features that are spuriously correlated with the class labels. To identify the key causal features, we propose a novel framework, COSDA, with susceptibility risk. COSDA leverages counterfactual tools from causal inference to measure a model’s susceptibility to input features by perturbing them and observing the resulting changes in predictions. Furthermore, COSDA is constructed from the perspectives of the optimizability, computability, and identifiability of susceptibility. To facilitate further research and practical use, we release an open-source repository containing our implementation and detailed configuration settings. This allows other researchers and practitioners to readily apply COSDA to their own open-set domain adaptation problems or related scenarios.