Aggregation in conformal e-classification

TL;DR

This paper explores methods for aggregating conformal e-predictors, focusing on experimental analysis of cross-conformal e-prediction and proposing simpler, more flexible modifications to improve efficiency and validity.

Contribution

It introduces simplified and more flexible modifications to cross-conformal e-prediction for better aggregation of conformal predictors.

Findings

Experimental validation of aggregation methods

Proposed modifications improve flexibility

Aggregation retains validity approximately

Abstract

Aggregating conformal predictors is a standard way of balancing their predictive and computational efficiency while retaining their validity, at least approximately. An important advantage of conformal e-predictors is that they are easier to aggregate without sacrificing their validity. This paper studies experimentally cross-conformal e-prediction, which is an existing method of aggregating conformal e-predictors, and its modifications that are conceptually simpler and more flexible.