Exact Optimization of Conformal Predictors via Incremental and Decremental Learning

Authors: Giovanni Cherubin, Konstantinos Chatzikokolakis, Martin Jaggi

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our findings empirically, and discuss when methods are suitable for CP optimization. and We empirically compare our techniques with i) original implementations of full CP, and ii) the most computationally efficient CP modification, ICP.
Researcher Affiliation Academia 1Alan Turing Institute, London, UK 2University of Athens 3EPFL.
Pseudocode Yes Algorithm 1 CP: computing a p-value for (x, ˆy)
Open Source Code Yes Code to reproduce the experiments: https://github. com/gchers/exact-cp-optimization.
Open Datasets Yes (In Appendix G, we further compare CP and ICP on the MNIST dataset.)
Dataset Splits No The paper uses generated data for its main experiments and does not specify explicit train/validation/test splits, percentages, or cross-validation methodology for its evaluation.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments.
Software Dependencies No The paper mentions using the 'scikit-learn library' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes For every training size n, chosen in the space [10, 105], we train the CP with a nonconformity measure, and use it to predict 100 test points. We set a timeout of 10 hours... We generate data for a binary classification problem with 30 features, by using the make classification() routine of the scikit-learn library. and We fix t/n = 0.5.