Time-uniform confidence bands for the CDF under nonstationarity

Authors: Paul Mineiro, Steven Howard

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

Reproducibility Variable Result LLM Response
Research Type Experimental We exhibit our techniques in various simulations in Section 4. These simulations explore the empirical behaviour of Algorithm 1 and Algorithm 2 when instantiated with ϵpdq 2d and curved boundary oracles Λ and Ξ.
Researcher Affiliation Industry Paul Mineiro Microsoft Research pmineiro@microsoft.com Steve Howard The Voleon Group steve@stevehoward.org
Pseudocode Yes Algorithm 1 Unit Interval Upper Bound. and Algorithm 2 Unit Interval Lower Bound. and Algorithm 3 Entire Real Line Upper Bound.
Open Source Code Yes Reference implementations which reproduce the figures are available at https://github.com/microsoft/csrobust.
Open Datasets No The paper uses synthetic data generated from distributions such as 'i.i.d. Beta(6, 3)', 'i.i.d. Log Normal(0, 1)', and 'i.i.d. Gaussian(0, 1)', but does not provide access information or citations for any publicly available or open dataset.
Dataset Splits No The paper describes using simulations and realizations from distributions, but it does not specify explicit training, validation, or test dataset splits. The work focuses on sequential methods where data accumulates over time.
Hardware Specification No The paper does not provide specific details regarding the hardware (e.g., GPU/CPU models, memory specifications) used to run its simulations or experiments.
Software Dependencies No The paper refers to Python implementations in its appendix (e.g., '.ipynb' files), but it does not list specific software dependencies or libraries with their version numbers required for replication.
Experiment Setup Yes For the proof we introduce an integer parameter η ě 2 which controls both the grid spacing (ϵpdq ηd) and the allocation of error probabilities to levels (δd α{pηdϵpdqq). In the main paper we set η 2. Our (synthetic) data set consists of five values... We use resolution ϵpdq 2d and coverage error α 1{20.