Robust Survival Analysis with Adversarial Regularization

Survival Analysis (SA) models the time until an event occurs, with applications in fields like medicine, defense, finance, and aerospace. Recent work shows that Neural Networks (NNs) can capture complex relationships in SA. However, dataset uncertainties (e.g., noisy measurements, human error) can degrade model performance. To address this, we leverage NN verification advances to create algorithms for robust, fully-parametric survival models. We introduce a robust loss function and use CROWN-IBP regularization to handle computational challenges in the Min-Max problem. Evaluating our approach on SurvSet datasets, we find that our Survival Analysis with Adversarial Regularization (SAWAR) method consistently outperforms baselines under various perturbations with respect to Negative Log Likelihood (NegLL), Integrated Brier Score (IBS), and Concordance Index (CI). This demonstrates that adversarial regularization enhances SA performance and calibration, mitigating data uncertainty and improving generalization across diverse datasets up to 150% across all perturbation magnitudes.