One-Shot Federated Conformal Prediction

Authors: Pierre Humbert, Batiste Le Bars, Aurélien Bellet, Sylvain Arlot

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate Fed CP-QQ on synthetic and real regression data sets. Additional experiments on unbalanced data sets and on Fed CP2-QQ are presented in Appendices A.1 and B.2.
Researcher Affiliation Academia 1Universit e Paris-Saclay, CNRS, Inria, Laboratoire de math ematiques d Orsay, 91405, Orsay, France. 2Universit e Lille, Inria, CNRS, Centrale Lille, UMR 9189, CRISt AL, F-59000 Lille.
Pseudocode Yes Algorithm 1 Fed CP-QQ
Open Source Code Yes The code of our two methods is available at https://github.com/pierre Hmbt/Fed CP-QQ.
Open Datasets Yes We evaluate our method on five public-domain regression data sets also considered by Romano et al. (2019) and Sesia & Romano (2021): physicochemical properties of protein tertiary structure (bio) (Rana, 2013); bike sharing (bike) (Fanaee-T & Gama, 2013); communities and crimes (community) (Redmond, 2011); Tennessee s student teacher achievement ratio (star) (Achilles et al., 2008); and concrete compressive strength (concrete) (Yeh, 1998).
Dataset Splits Yes For each experiment, we split the full data set into three parts: a training set (40%), a calibration set (40%), and a test set (20%).
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions software components like 'quantile regression forests' and 'Neural Networks for quantile regression', but does not provide specific version numbers for software dependencies or libraries.
Experiment Setup Yes The number of trees in the forest is set to 1000, the two parameters controlling the coverage rate on the training data are tuned using cross-validation and the remaining hyperparameters are set as done by Romano et al. (2019). ... CQR with Neural Networks (NN) for quantile regression (Taylor, 2000) the architecture and the parameters are those used by Romano et al. (2019).