Target Conditioned Representation Independence (TCRI); From Domain-Invariant to Domain-General Representations
This addresses domain generalization for machine learning models, but it appears incremental as it builds on existing methods with more complete constraints.
The paper tackles domain generalization by proposing the Target Conditioned Representation Independence (TCRI) objective, which uses conditional independence constraints to learn invariant mechanisms, resulting in competitive average accuracy and improved worst-domain accuracy on synthetic and real-world data.
We propose a Target Conditioned Representation Independence (TCRI) objective for domain generalization. TCRI addresses the limitations of existing domain generalization methods due to incomplete constraints. Specifically, TCRI implements regularizers motivated by conditional independence constraints that are sufficient to strictly learn complete sets of invariant mechanisms, which we show are necessary and sufficient for domain generalization. Empirically, we show that TCRI is effective on both synthetic and real-world data. TCRI is competitive with baselines in average accuracy while outperforming them in worst-domain accuracy, indicating desired cross-domain stability.