Aggregated Conformal Predictor

Inductive conformal predictors are a computationally efficient version of conformal predictors, but they can be argued to be informationally inefficient. Aggregated conformal predictors were introduced by Carlsson et al. (2014) in an attempt to achieve both computational and informational efficiency; however, they lose the formal property of validity. One version of aggregated conformal predictors is cross-conformal predictors (Vovk, 2015). An empirical study of the validity of aggregated conformal predictors was carried out by Linusson et al. (2017).

References

• Lars Carlsson, Martin Eklund, and Ulf Norinder (2014). Aggregated Conformal Prediction. In: COPA 2014. Proceedings of the IFIP International Conference on Artificial Intelligence Applications and Innovations (AIAI 2014), pp. 231 - 240.
• Henrik Linusson, Ulf Norinder, Henrik Boström, Ulf Johansson, Tuve Löfström (2017). On the Calibration of Aggregated Conformal Predictors. Proceedings of Machine Learning Research 60:154 - 173.
• Vladimir Vovk (2015). Cross-conformal predictors. Annals of Mathematics and Artificial Intelligence (Special Issue on Conformal Prediction and its Applications) 74:9 - 28.