Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

Statistical Parity with Exponential Weights

Authors: Stephen Pasteris, Chris Hicks, Vasilios Mavroudis

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical Specifically, we present a meta-algorithm that transforms any efficient implementation of Hedge (or, equivalently, any discrete Bayesian inference algorithm) into an efficient contextual bandit algorithm that guarantees exact statistical parity on every trial. Compared to any comparator that satisfies the same statistical parity constraint, the algorithm achieves the same asymptotic regret bound as running the equivalent instance of Exp4 for each group. We also address the scenario where the target parity distribution is unknown and must be estimated online. Finally, using online-to-batch conversion, we extend our approach to the batch classification setting.
Researcher Affiliation Academia Stephen Pasteris The Alan Turing Institute London UK EMAIL Chris Hicks The Alan Turing Institute London UK EMAIL Vasilios Mavroudis The Alan Turing Institute London UK EMAIL
Pseudocode Yes 4.3 Pseudocode We now give the pseudocode of SPEW. On any trial t [T] SPEW does the following. Receive µt, xt and ct for c C do ξt(c, , ) QUERY[c] for a A do x Gt(c) µt(c, x)ξt(c, x, a) end for end for for (c, a) C A do δt(c, a) maxc C ωt(c , a) ωt(c, a) end for βt P a A maxc C δt(c, a) for a A do ψ t(a) (ξt(ct, xt, a) + δt(ct, a))/(1 + βt) end for ht 1 P a A ψ t(a) for a A do π t(a) ψ t(a) + ht/K end for Draw at from probability distribution π t Receive ℓt(at) for a A do κt(a) argmaxc C ωt(c, a) κ t(a) argminc C ωt(c, a) end for for c C do for x Gt(c) do λt(c, x, a) Jβt 1KJ(c, x, a) = (ct, xt, at)Kℓt(at)/π t(at) + Jc = κt(a)Kµt(c, x) Jc = κ t(a)Kµt(c, x) end for UPDATE[c](x, λt(c, x, )) end for end for
Open Source Code No No explicit statement about providing open-source code for the methodology described in this paper was found. The NeurIPS checklist for question 5 also states 'NA' and 'No experiments' as justification.
Open Datasets No No explicit mention of specific publicly available or open datasets used for empirical evaluation was found. The NeurIPS checklist for question 5 also states 'NA' and 'No experiments' as justification.
Dataset Splits No No specific dataset split information is provided, as the paper focuses on theoretical contributions and does not conduct experiments with datasets. The NeurIPS checklist for question 6 also states 'NA' and 'No experiments' as justification.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory amounts) are provided, as the paper is theoretical and does not involve running experiments requiring such specifications. The NeurIPS checklist for question 8 also states 'NA' and 'No experiments' as justification.
Software Dependencies No The paper mentions general algorithms like 'Hedge' and 'BELIEFPROPAGATION [29]' but does not list specific software dependencies with version numbers for experimental reproduction. The NeurIPS checklist for question 4 also states 'NA' and 'No experiments' as justification.
Experiment Setup No No specific experimental setup details (e.g., hyperparameter values, training configurations) are provided, as the paper is theoretical and does not involve running experiments. The NeurIPS checklist for question 6 also states 'NA' and 'No experiments' as justification.