Multi-Source Survival Domain Adaptation

Survival analysis is the branch of statistics that studies relation between characteristics living entities and their respective survival times, taking into account partial information held by censored cases. A good can, for example, determine whether one medical treatment a group patients better than another. With rise machine learning, can be modeled as learning function maps studied to times. To succeed with that, there are three crucial issues tackled. First, some patient data censored: we do not know true times all patients. Second, scarce, which led past research treat different illness types domains in multi-task setup. Third, need adaptation new or extremely rare types, where little no labels available. In contrast previous setups, want investigate how efficiently adapt target domain from multiple source domains. For this, introduce metric corresponding discrepancy measure distributions. These allow us define while incorporating data, would otherwise have dropped. Our experiments on two cancer sets reveal superb performance domains, recommendation, weight matrix plausible explanation.