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..
Online Multi-Class Selection with Group Fairness Guarantee
Authors: Faraz Zargari, Hossein Jazi, Lyndon Hallett, Bo Sun, Xiaoqi Tan
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we evaluate our algorithms to investigate the empirical fairness and efficiency under different fairness notions and algorithms. We also explore how untrusted black-box advice can help the improvement of fairness. |
| Researcher Affiliation | Academia | Faraz Zargari University of Alberta EMAIL Hossein Nekouyan University of Alberta EMAIL Lyndon Hallett University of Alberta EMAIL Bo Sun University of Ottawa, Vector Institute EMAIL Xiaoqi Tan University of Alberta EMAIL |
| Pseudocode | Yes | Algorithm 1: Randomized Set-Aside with GFQ guarantee (R-SETASIDE-GFQ) Input: B; {mj, θj} j [K] ... Algorithm 2: Lossless Online Rounding (ROUNDING) Input: κ, zn, zp, x ... Algorithm 3: Randomized Set-Aside with β-PF guarantee (R-SETASIDE-PF) Input: B, {θj} j [K]. Initialize: Sale unit index κ1 = 1, utilization levels {zi,j 0 = 0} i,j [K], z G 0 = 0 and z0 = 0. ... Algorithm 4: Deterministic Set-Aside with GFQ guarantee (D-SETASIDE-GFQ) Input: B; {mj} j [K]; {λ i } i [B M]. Initialize: Unit index κ1 = 1, {κj 1 = 1} j [K]. ... Algorithm 5: Fractional OMc S with GFQ guarantee (FRAC-GFQ) Input: (mj, θj), j [K]. Initialization: Initial global utilization, u0 = 0; Initial utilization of class j, uj 0 = 0, j [K]. |
| Open Source Code | No | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [No] Justification: While we did not provide open access to our code, we clearly state in the paper that the dataset used in our experiments is publicly available and properly cited in the main body. Moreover, as a theory-oriented paper, our code is limited to direct implementations of the proposed algorithms without any additional tuning or tweaks, making the experiments straightforward to reproduce based on the descriptions provided. |
| Open Datasets | Yes | We evaluate our theoretical results using the Google Cluster Data [Goo15], which records CPU usage over time for three request types and here we consider them as distinct classes. |
| Dataset Splits | No | The normalized CPU allocations are scaled to fit our model, and requests needing more than one CPU units are split into single-unit requests. We assign valuations randomly with θ1 = 5, θ2 = 10, and θ3 = 15, with B = 100. |
| Hardware Specification | No | Question: For each experiment, does the paper provide sufficient information on the computer resources (type of compute workers, memory, time of execution) needed to reproduce the experiments? Answer: [No] Justification: As a theory-focused paper, our experiments are solely intended to validate the correctness and behavior of the proposed algorithms. They do not require any specific computational configuration, and therefore details such as compute resources, memory, or execution time are not necessary for reproduction. |
| Software Dependencies | No | The paper does not explicitly mention any specific software dependencies or their version numbers used for the experimental setup. It focuses on the algorithmic design and theoretical analysis. |
| Experiment Setup | Yes | We assign valuations randomly with θ1 = 5, θ2 = 10, and θ3 = 15, with B = 100. We analyze OMc S with GFQ under deterministic (D-SETASIDE-GFQ) and randomized (R-SETASIDE-GFQ) algorithms, denoted as d-GFQ and r-GFQ respectively. For our analysis, we set mj = 5 for each class j [K]. We also study OMc S with β-proportional fairness denoted as β-PF (R-SETASIDE-PF with b = 1) and without fairness consideration denoted as α-CR (R-SETASIDE-PF with b = 1). To model the advice, we follow [LCS+24] approach. Let ξ [0, 1] represents an adversarial probability. When ξ = 0, ADV provides the optimal solution, and when ξ = 1, ADV is fully adversarial. Formally, let {x t : t [T]} denote the optimal decisions and {ˇxt : t [T]} the decisions that minimize the objective. Then, in expectation, the advised decisions are given by ADV = {(1 ξ)x t + ξˇxt : t [T]}. Under this advice model, we examine Li LA with GFQ (GFQ-LA) and β-proportional fairness (β-PF-LA) as well. In our first experiment, we set ξ = 0 and ϵ = 1.25. |