The article ‘High-Dimensional Prediction for Sequential Decision Making’ by Georgy Noarov et al. explores the problem of making predictions for a high-dimensional state chosen by an adversary. The predictions are unbiased and subject to a collection of conditioning events, aimed at assisting downstream decision makers. The researchers provide efficient algorithms for this problem, and highlight applications arising from selecting an appropriate set of conditioning events. They also delve into the realm of uncertainty quantification in machine learning, discussing the production of prediction sets for online adversarial multiclass and multilabel classification. The algorithms provide transparent coverage guarantees, which imply strong online adversarial conditional validity guarantees for downstream prediction set algorithms.
Publication date: 27 Oct 2023
Project Page: https://arxiv.org/abs/2310.17651v1
Paper: https://arxiv.org/pdf/2310.17651
