Generalized Fast Exact Conformalization

Authors: Diyang Li

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide experimental results on real-world benchmarks to validate our derived algorithm.The significant speedups of our algorithm as compared to the existing standard methods are demonstrated across numerous benchmarks.
Researcher Affiliation Academia Diyang Li Cornell University diyang01@cs.cornell.edu
Pseudocode Yes Algorithm 1 Fast Exact Conformalization Algorithm
Open Source Code No This paper does not release new assets.
Open Datasets Yes Our experiments were conducted using real-world datasets. We employ real-world datasets from Open ML [29] and UCI repository [30] in simulations.
Dataset Splits Yes We randomly partition the dataset into training set, testing set, and calibration set (used in SCP) with 70%, 10%, and 20% of the total samples.
Hardware Specification Yes All experiments presented in this study were conducted on a workstation running the Ubuntu 18.04 operating system, equipped with Intel Xeon Gold 5218R CPU 64 and 60.9 GB of RAM.
Software Dependencies No We integrate a system of ordinary differential equations using lsoda from the FORTRAN library, where an interface for Sci Py is available using the odepack.
Experiment Setup Yes The concrete parameter settings of ODE solver are shown in the Table 3, wherein the numerical solver exploit the Runge-Kutta method of order 4 or 5.We use the conformity score function Ai = |yi ηw (xi)|.