A1138
Title: Flow-based conditional predictive inference
Authors: Youhui Ye - Virginia Tech (United States)
Meimei Liu - Virginia Tech (United States)
Xin Xing - Virginia Tech University (United States) [presenting]
Abstract: The objective of predictive inference is to determine precise levels of confidence in predictions for new objects using past experience. A novel method called Flow-based Conditional Predictive Inference (FCI) is introduced for building predictive sets for complex and high-dimensional data. FCI uses ideas from adversarial flow to transfer input data to a random vector with known distributions, allowing for the construction of a p-value for uncertainty quantification. Our approach is applicable and robust, even when the testing data is contaminated. The method, robust flow-based conformal inference, on benchmark datasets is evaluated, and it is demonstrated that it produces effective predictive sets and accurate outlier detection, outperforming other approaches in terms of power.
