Jae-Kwang Kim – Weight calibration for causal inference and transfer learning
B232 Seminar Room, IBS 55 Expo-ro Yuseong-gu, Daejeon, Daejeon, Korea, Republic ofAbstract: Weight calibration is a popular technique in handling covariate-shift problem in causal inference. It can be viewed as a dual optimization problem for incorporating the implicit regression model. We introduce the generalized entropy calibration as a general tool for weight calibration. Several interesting applications will be introduced in the context of causal inference. Furthermore, weight calibration can be used to transfer learning, which combines information from two different samples, one for source data and the other for target data.