pith. machine review for the scientific record. sign in

arxiv: 2605.06656 · v1 · submitted 2026-05-07 · 💻 cs.LG · cs.DM· cs.ET· math.OC

Recognition: unknown

Why Global LLM Leaderboards Are Misleading: Small Portfolios for Heterogeneous Supervised ML

Jai Moondra , Ayela Chughtai , Bhargavi Lanka , Swati Gupta
Authors on Pith no claims yet

Pith reviewed 2026-05-08 12:01 UTC · model grok-4.3

classification 💻 cs.LG cs.DMcs.ETmath.OC
keywords LLM leaderboardsBradley-Terry rankingheterogeneous preferenceslanguage subgroupsportfoliosset coverELO scoressupervised ML
0
0 comments X

The pith

Global Bradley-Terry rankings of LLMs are misleading because nearly two-thirds of decisive votes cancel out under structured heterogeneity.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that a single global ranking from pairwise human feedback on LLMs conceals strong disagreements that vary systematically by language, task, and time. Analysis of 89,000 comparisons across 52 models shows that top-50 models have pairwise win probabilities no higher than 0.53 and that language-based grouping produces ELO spreads two orders of magnitude larger than the global ranking. To manage this heterogeneity in supervised machine learning, the work defines small (λ, ν)-portfolios as sets of models that keep error at most λ while covering at least ν fraction of users, solved via set-cover with VC-dimension bounds. On the Arena data these portfolios recover five distinct rankings that together cover more than 96 percent of votes, versus 21 percent for the global ranking.

Core claim

The best-fit global BT ranking is misleading: nearly 2/3 of decisive votes cancel out, top-50 models are statistically indistinguishable with pairwise win probabilities ≤0.53, and language-driven grouping yields two orders of magnitude higher ELO spread. (λ, ν)-portfolios recover five distinct BT rankings covering >96% of votes versus 21% for the global ranking, and a portfolio of six LLMs covers twice as many votes as the global top-six.

What carries the argument

(λ, ν)-portfolios: small sets of models that achieve prediction error at most λ while covering at least ν fraction of users, formulated as a set-cover problem with VC-dimension guarantees.

If this is right

  • Leaderboards should report multiple small portfolios rather than one global ranking to reflect coherent subpopulations.
  • A portfolio of six LLMs covers twice the votes of the global top-six, showing that modest sets suffice for high coverage.
  • The same portfolio construction applied to fairness-regularized classifiers on COMPAS data identifies blind spots in the dataset.
  • Language families produce internally consistent but mutually conflicting rankings, so global aggregation erases most signal.
  • Five recovered rankings together account for over 96 percent of observed votes, demonstrating that the heterogeneity is low-dimensional.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If language remains the dominant grouping factor on new tasks, the same five portfolios could be reused across creative writing, coding, and reasoning benchmarks.
  • Policymakers could run the portfolio construction on fairness-sensitive datasets to surface which subpopulations are poorly served by any single model.
  • The set-cover formulation with VC guarantees may extend directly to other heterogeneous ranking settings such as product recommendations or clinical treatment selection.
  • Tracking how the recovered portfolios change when new languages or tasks are added would test the claimed stability without assuming the current five suffice forever.

Load-bearing premise

The heterogeneity in preferences is structured enough, primarily by language, and stable enough that the recovered portfolios remain effective for new users and tasks outside the Arena sample.

What would settle it

Collect a fresh set of pairwise comparisons from a different user population or new tasks and check whether the five portfolios still cover over 90 percent of votes or whether the global ranking covers a comparable fraction.

Figures

Figures reproduced from arXiv: 2605.06656 by Ayela Chughtai, Bhargavi Lanka, Jai Moondra, Swati Gupta.

Figure 1
Figure 1. Figure 1: Model differentiation across ranking types. Each point is a BT ranking fitted on a group view at source ↗
Figure 2
Figure 2. Figure 2: In-group predictive performance of 388 subpopulation BT rankings vs. the global ranking. view at source ↗
Figure 3
Figure 3. Figure 3: Language-family BT rankings capture distinct preference structure beyond the global view at source ↗
Figure 4
Figure 4. Figure 4: Coverage by global ranking and by our portfolios. view at source ↗
Figure 5
Figure 5. Figure 5: Heterogeneity in LLM rankings across languages. Rows: 15 languages with the most votes view at source ↗
Figure 6
Figure 6. Figure 6: Distribution of pairwise win probabilities view at source ↗
Figure 7
Figure 7. Figure 7: Language Family BT rankings achieve better coverage than a global BT ranking. The top view at source ↗
Figure 8
Figure 8. Figure 8: Coverage of the global BT ranking by language at view at source ↗
Figure 9
Figure 9. Figure 9: Per-vote portfolio error distributions for MIP and greedy BT ranking portfolios at view at source ↗
Figure 10
Figure 10. Figure 10: Minimum number of models needed to achieve target population coverage across view at source ↗
read the original abstract

Ranking LLMs via pairwise human feedback underpins current leaderboards for open-ended tasks, such as creative writing and problem-solving. We analyze ~89K comparisons in 116 languages from 52 LLMs from Arena, and show that the best-fit global Bradley-Terry (BT) ranking is misleading. Nearly 2/3 of the decisive votes cancel out, and even the top 50 models according to the global BT ranking are statistically indistinguishable (pairwise win probabilities are at most 0.53 within the top 50 models). We trace this failure to strong, structured heterogeneity of opinions across language, task, and time. Moreover, we find an important characteristic - *language* plays a key role. Grouping by language (and families) increases the agreement of votes massively, resulting in two orders of magnitude higher spread in the ELO scores (i.e., very consistent rankings). What appears as global noise is in fact a mixture of coherent but conflicting subpopulations. To address such heterogeneity in supervised machine learning, we introduce the framework of $(\lambda, \nu)$-portfolios, which are small sets of models that achieve a prediction error at most $\lambda$, "covering" at least a $\nu$ fraction of users. We formulate this as a variant of the set cover problem and provide guarantees using the VC dimension of the underlying set system. On the Arena data, our algorithms recover just 5 distinct BT rankings that cover over 96% of votes at a modest $\lambda$, compared to the 21% coverage by the global ranking. We also provide a portfolio of 6 LLMs that cover twice as many votes as the top-6 LLMs from a global ranking. We further construct portfolios for a classification problem on the COMPAS dataset using an ensemble of fairness-regularized classification models and show that these portfolios can be used to detect blind spots in the data, which might be of independent interest to policymakers.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript analyzes ~89K pairwise human comparisons from the LMSYS Chatbot Arena across 52 LLMs and 116 languages. It shows that the global Bradley-Terry ranking is misleading because nearly two-thirds of decisive votes cancel out and the top-50 models are statistically indistinguishable (pairwise win probabilities ≤0.53). The authors attribute this to structured heterogeneity, with language playing a key role in increasing ELO spread by two orders of magnitude. They introduce (λ, ν)-portfolios as small sets of models achieving prediction error at most λ while covering at least a ν fraction of users, formulated as a variant of the set-cover problem with VC-dimension guarantees. On the Arena data, 5 such portfolios cover over 96% of votes compared to 21% for the global ranking; a 6-model portfolio covers twice as many votes as the global top-6. The framework is also applied to the COMPAS dataset to detect data blind spots via fairness-regularized ensembles.

Significance. If the empirical claims on vote cancellation and coverage hold and the VC-dimension analysis is tightened, the work provides a concrete demonstration that global leaderboards obscure coherent subpopulations and offers a set-cover-based framework for constructing small, high-coverage model portfolios in heterogeneous supervised ML. Strengths include the use of large-scale public Arena data, the independent set-cover reduction (not circular with the BT fit), and the dual application to LLM evaluation and fairness auditing on COMPAS. These elements could influence leaderboard design and user-model matching if the language-driven structure generalizes.

major comments (3)
  1. [§3] §3 (Arena data analysis): The Bradley-Terry fitting procedure is not specified in detail, including the optimization algorithm, tie handling, or any regularization. This is load-bearing for the central claims of ~2/3 vote cancellation and statistical indistinguishability of the top-50 models (pairwise win probabilities ≤0.53), as these quantities depend directly on the fitted parameters.
  2. [§4.2] §4.2 (set-cover formulation and VC-dimension guarantees): The paper invokes VC-dimension bounds for the (λ, ν)-portfolio set system but does not instantiate or bound the VC dimension for the empirical 89K-vote Arena data (or the COMPAS application). Without this, it is impossible to assess whether the theoretical guarantees support the reported 96% coverage or are merely existential.
  3. [§5] §5 (portfolio evaluation and generalization): The claim that 5 BT rankings cover >96% of votes (vs. 21% global) and that a 6-model portfolio covers twice as many votes as the global top-6 is purely in-sample. No temporal, task, or user hold-out validation is provided to test whether the recovered portfolios remain effective on new data, which is required to substantiate the utility for new users/tasks stated in the introduction and conclusion.
minor comments (2)
  1. [§4.1] The definition of the (λ, ν)-portfolio in §4.1 uses 'prediction error at most λ' without clarifying whether this is 0-1 loss, cross-entropy, or another metric, and how it is estimated from the pairwise votes.
  2. Table or figure reporting the exact λ and ν values used for the 5-portfolio recovery and the 6-model portfolio would improve clarity and allow direct comparison to the global baseline.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below, providing clarifications on our methods and indicating where we will revise the manuscript for greater rigor and transparency.

read point-by-point responses

  1. Referee: [§3] §3 (Arena data analysis): The Bradley-Terry fitting procedure is not specified in detail, including the optimization algorithm, tie handling, or any regularization. This is load-bearing for the central claims of ~2/3 vote cancellation and statistical indistinguishability of the top-50 models (pairwise win probabilities ≤0.53), as these quantities depend directly on the fitted parameters.

    Authors: We agree that the Bradley-Terry fitting details require expansion for reproducibility. In the revised manuscript we will add a dedicated paragraph in §3 specifying that we fit the model via maximum likelihood estimation using the iterative Luce spectral ranking algorithm (equivalent to logistic regression on the pairwise outcomes), that ties are excluded from the likelihood as non-decisive, and that no parameter regularization is applied. These choices directly produce the reported ~2/3 cancellation rate and the pairwise win probabilities ≤0.53 within the top 50. The associated code will be released with the revision. revision: yes

  2. Referee: [§4.2] §4.2 (set-cover formulation and VC-dimension guarantees): The paper invokes VC-dimension bounds for the (λ, ν)-portfolio set system but does not instantiate or bound the VC dimension for the empirical 89K-vote Arena data (or the COMPAS application). Without this, it is impossible to assess whether the theoretical guarantees support the reported 96% coverage or are merely existential.

    Authors: The VC-dimension analysis supplies a general sample-complexity guarantee for the existence of small (λ, ν)-portfolios under the set-cover formulation; it is not intended to predict or bound the observed in-sample coverage. The 96% figure is obtained by direct application of the greedy algorithm to the 89K observed votes. In the revision we will explicitly distinguish these roles in §4.2, state that the hypothesis class has VC dimension at most 52 (the number of distinct LLMs), and note that this loose bound is not used to compute the empirical coverage. The same clarification will be added for the COMPAS application. revision: partial

  3. Referee: [§5] §5 (portfolio evaluation and generalization): The claim that 5 BT rankings cover >96% of votes (vs. 21% global) and that a 6-model portfolio covers twice as many votes as the global top-6 is purely in-sample. No temporal, task, or user hold-out validation is provided to test whether the recovered portfolios remain effective on new data, which is required to substantiate the utility for new users/tasks stated in the introduction and conclusion.

    Authors: We acknowledge that all reported coverage numbers are in-sample. The manuscript’s central contribution is to exhibit the existence of small, high-coverage portfolios that reveal structured heterogeneity within the given Arena snapshot. We agree that temporal or user-level hold-out validation would strengthen claims about utility for future users. The public Arena release lacks reliable user identifiers and clean temporal splits, precluding straightforward out-of-sample tests without new data collection. In the revision we will add an explicit limitations paragraph in §5 and the conclusion that states this constraint and outlines how streaming Arena data could be used for future temporal validation. revision: partial

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper's central results are empirical computations performed on the external Arena dataset (~89K votes) using standard Bradley-Terry fitting, followed by direct measurement of vote coverage percentages (21% global vs. 96% for the recovered 5 rankings). The (λ, ν)-portfolio framework is introduced as a new set-cover formulation whose VC-dimension guarantees are drawn from standard learning theory rather than self-citation or data-specific fitting. No claimed quantity (ELO spreads, coverage figures, or portfolio sizes) is defined in terms of itself or recovered by construction from a fitted parameter; the COMPAS application is likewise an independent empirical demonstration. The analysis therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 1 invented entities

The central claims rest on the existence of structured heterogeneity capturable by language groups and on the applicability of VC-dimension set-cover bounds to the induced set system; λ and ν are user-chosen thresholds rather than fitted constants.

free parameters (2)
  • λ
    Error tolerance threshold for acceptable model performance within a portfolio; chosen by the user rather than fitted to data.
  • ν
    Minimum coverage fraction of users that the portfolio must serve; chosen by the user.
axioms (1)
  • domain assumption The collection of user-model sets has finite VC dimension
    Invoked to obtain polynomial-time approximation guarantees for the set-cover formulation of portfolio selection.
invented entities (1)
  • (λ, ν)-portfolio no independent evidence
    purpose: A small set of models that together achieve bounded error for at least fraction ν of users
    New construct introduced to replace single global rankings when preferences are heterogeneous.

pith-pipeline@v0.9.0 · 5679 in / 1641 out tokens · 84801 ms · 2026-05-08T12:01:14.471552+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

299 extracted references · 169 canonical work pages · 1 internal anchor

  1. [1]

    2018 , url =

    Constructing. SIAM Journal on Computing , author =. doi:10.1137/16M1061850 , abstract =

    work page doi:10.1137/16m1061850

  2. [2]

    Journal of the Operational Research Society , author =

    Optimal portfolios using linear programming models , volume =. Journal of the Operational Research Society , author =. 2004 , pages =. doi:10.1057/palgrave.jors.2601765 , abstract =

    work page doi:10.1057/palgrave.jors.2601765 2004

  3. [3]

    Zhuo, Zhengjia and Nagarajan, Viswanath , month = may, year =. A. doi:10.48550/arXiv.2505.15641 , abstract =

    work page doi:10.48550/arxiv.2505.15641

  4. [4]

    Probability

    Hoeffding, Wassily , editor =. Probability. The. 1994 , doi =

    1994

  5. [5]

    Chen, Violet (Xinying) and Hooker, J. N. , month = feb, year =. A. Proceedings of the. doi:10.1145/3375627.3375844 , abstract =

    work page doi:10.1145/3375627.3375844

  6. [6]

    Journal of Algorithms , author =

    Bicriteria. Journal of Algorithms , author =. 1998 , keywords =. doi:10.1006/jagm.1998.0930 , abstract =

    work page doi:10.1006/jagm.1998.0930 1998

  7. [7]

    Byrka, Jarosław and Fleszar, Krzysztof and Rybicki, Bartosz and Spoerhase, Joachim , month = dec, year =. Bi-. Proceedings of the 2015. doi:10.1137/1.9781611973730.49 , pages =

    work page doi:10.1137/1.9781611973730.49 2015

  8. [8]

    Management Science , author =

    An. Management Science , author =. 1984 , keywords =. doi:10.1287/mnsc.30.11.1268 , abstract =

    work page doi:10.1287/mnsc.30.11.1268 1984

  9. [9]

    Operations Research , author =

    Diversity. Operations Research , author =. 2008 , keywords =. doi:10.1287/opre.1070.0413 , abstract =

    work page doi:10.1287/opre.1070.0413 2008

  10. [10]

    Mathematical Programming , author =

    New approaches to multi-objective optimization , volume =. Mathematical Programming , author =. 2014 , keywords =. doi:10.1007/s10107-013-0703-7 , abstract =

    work page doi:10.1007/s10107-013-0703-7 2014

  11. [11]

    ACM Comput

    Evolutionary. ACM Comput. Surv. , author =. 2021 , pages =. doi:10.1145/3470971 , abstract =

    work page doi:10.1145/3470971 2021

  12. [12]

    A theory of fairness and social welfare , isbn =

    Fleurbaey, Marc and Maniquet, François , month = jan, year =. A theory of fairness and social welfare , isbn =

  13. [13]

    INFORMS Journal on Computing , author =

    Approximation. INFORMS Journal on Computing , author =. 2021 , keywords =. doi:10.1287/ijoc.2020.1028 , abstract =

    work page doi:10.1287/ijoc.2020.1028 2021

  14. [14]

    Approximability and

    Glaßer, Christian and Reitwießner, Christian and Schmitz, Heinz and Witek, Maximilian , editor =. Approximability and. Programs,. 2010 , keywords =. doi:10.1007/978-3-642-13962-8_20 , abstract =

    work page doi:10.1007/978-3-642-13962-8_20 2010

  15. [15]

    Theory of Computing Systems , author =

    The. Theory of Computing Systems , author =. 2022 , note =. doi:10.1007/s00224-021-10066-5 , abstract =

    work page doi:10.1007/s00224-021-10066-5 2022

  16. [16]

    Operations Research , author =

    Risk-. Operations Research , author =. 2018 , keywords =. doi:10.1287/opre.2018.1729 , abstract =

    work page doi:10.1287/opre.2018.1729 2018

  17. [17]

    The Review of Economic Studies , author =

    When. The Review of Economic Studies , author =. 2023 , pages =. doi:10.1093/restud/rdac050 , abstract =

    work page doi:10.1093/restud/rdac050 2023

  18. [18]

    INFORMS Journal on Applied Analytics , author =

    Bus. INFORMS Journal on Applied Analytics , author =. 2020 , keywords =. doi:10.1287/inte.2019.1015 , abstract =

    work page doi:10.1287/inte.2019.1015 2020

  19. [19]

    Proceedings of the National Academy of Sciences , author =

    Optimizing schools’ start time and bus routes , volume =. Proceedings of the National Academy of Sciences , author =. 2019 , pages =. doi:10.1073/pnas.1811462116 , abstract =

    work page doi:10.1073/pnas.1811462116 2019

  20. [20]

    Operations Research , author =

    Reshaping. Operations Research , author =. doi:10.1287/opre.2022.0035 , abstract =

    work page doi:10.1287/opre.2022.0035 2022

  21. [21]

    Management Science , author =

    Multi-. Management Science , author =. 1977 , pages =. doi:10.1287/mnsc.23.7.703 , abstract =

    work page doi:10.1287/mnsc.23.7.703 1977

  22. [22]

    Esmaeili, Seyed and Brubach, Brian and Srinivasan, Aravind and Dickerson, John , year =. Fair. Advances in

  23. [23]

    European Journal of Operational Research , author =

    Price of fairness in two-agent single-machine scheduling problems , volume =. European Journal of Operational Research , author =. 2019 , keywords =. doi:10.1016/j.ejor.2018.12.048 , abstract =

    work page doi:10.1016/j.ejor.2018.12.048 2019

  24. [24]

    Chierichetti, Flavio and Kumar, Ravi and Lattanzi, Silvio and Vassilvitskii, Sergei , year =. Fair. Advances in

  25. [25]

    , month = jan, year =

    Menon, Aditya Krishna and Williamson, Robert C. , month = jan, year =. The cost of fairness in binary classification , url =. Proceedings of the 1st

  26. [26]

    European Journal of Operational Research , author =

    Price of. European Journal of Operational Research , author =. 2017 , keywords =. doi:10.1016/j.ejor.2016.08.013 , abstract =

    work page doi:10.1016/j.ejor.2016.08.013 2017

  27. [27]

    Theory of Computing Systems , author =

    The. Theory of Computing Systems , author =. 2012 , keywords =. doi:10.1007/s00224-011-9359-y , abstract =

    work page doi:10.1007/s00224-011-9359-y 2012

  28. [28]

    Operations Research , author =

    The. Operations Research , author =. 2011 , note =. doi:10.1287/opre.1100.0865 , abstract =

    work page doi:10.1287/opre.1100.0865 2011

  29. [29]

    The. Found. Trends Theor. Comput. Sci. , author =. 2014 , pages =. doi:10.1561/0400000042 , abstract =

    work page doi:10.1561/0400000042 2014

  30. [30]

    Efficient modality selection in multimodal learning , volume =. J. Mach. Learn. Res. , author =. 2024 , pages =

    2024

  31. [31]

    Rosa and N

    Unsettling race and language:. Language in Society , author =. 2017 , pages =. doi:10.1017/S0047404517000562 , abstract =

    work page doi:10.1017/s0047404517000562 2017

  32. [32]

    1997 , note =

    Handbook of autism and pervasive developmental disorders, 2nd ed , isbn =. 1997 , note =

    1997

  33. [33]

    Coates, Jennifer , month = dec, year =. Women,

  34. [34]

    JAMA Neurology , author =

    Assessment of. JAMA Neurology , author =. 2019 , pages =. doi:10.1001/jamaneurol.2018.4249 , abstract =

    work page doi:10.1001/jamaneurol.2018.4249 2019

  35. [35]

    National Institute on Aging , month = dec, year =

    Data shows racial disparities in. National Institute on Aging , month = dec, year =

  36. [36]

    Biomolecules , author =

    Symptomatic,. Biomolecules , author =. 2021 , keywords =. doi:10.3390/biom11111635 , abstract =

    work page doi:10.3390/biom11111635 2021

  37. [37]

    Neurology , author =

    Autism,. Neurology , author =. 2011 , pages =. doi:10.1212/WNL.0b013e3182166dc7 , abstract =

    work page doi:10.1212/wnl.0b013e3182166dc7 2011

  38. [38]

    American Journal of Epidemiology , author =

    Is the. American Journal of Epidemiology , author =. 2001 , pages =. doi:10.1093/aje/153.2.132 , abstract =

    work page doi:10.1093/aje/153.2.132 2001

  39. [39]

    Journal of Clinical and Experimental Neuropsychology , author =

    Greater cognitive deterioration in women than men with. Journal of Clinical and Experimental Neuropsychology , author =. 2012 , pmid =. doi:10.1080/13803395.2012.712676 , abstract =

    work page doi:10.1080/13803395.2012.712676 2012

  40. [40]

    Psychiatry Research , author =

    Speech and language patterns in autism:. Psychiatry Research , author =. 2024 , keywords =. doi:10.1016/j.psychres.2024.116109 , abstract =

    work page doi:10.1016/j.psychres.2024.116109 2024

  41. [41]

    Diagnostics , author =

    Artificial. Diagnostics , author =. 2021 , keywords =. doi:10.3390/diagnostics11081473 , abstract =

    work page doi:10.3390/diagnostics11081473 2021

  42. [42]

    Sensors , author =

    Revolutionizing the. Sensors , author =. 2023 , keywords =. doi:10.3390/s23094184 , abstract =

    work page doi:10.3390/s23094184 2023

  43. [43]

    Proceedings of the 2025 Annual ACM-SIAM Symposium on Discrete Algorithms (SODA) , author =

    Balancing. Proceedings of the 2025 Annual ACM-SIAM Symposium on Discrete Algorithms (SODA) , author =. 2025 , doi =

    2025

  44. [44]

    arXiv preprint arXiv:2502.09724 , year=

    Kim, Cheol Woo and Moondra, Jai and Verma, Shresth and Pollack, Madeleine and Kong, Lingkai and Tambe, Milind and Gupta, Swati , month = feb, year =. Navigating the. doi:10.48550/arXiv.2502.09724 , abstract =

    work page doi:10.48550/arxiv.2502.09724

  45. [45]

    Spectral sparsification of graphs: theory and algorithms , volume =. Commun. ACM , author =. 2013 , pages =. doi:10.1145/2492007.2492029 , abstract =

    work page doi:10.1145/2492007.2492029 2013

  46. [46]

    Physical Review X , author =

    Quantum. Physical Review X , author =. 2020 , pages =. doi:10.1103/PhysRevX.10.021067 , abstract =

    work page doi:10.1103/physrevx.10.021067 2020

  47. [47]

    Some optimal inapproximability results , volume =. J. ACM , author =. 2001 , pages =. doi:10.1145/502090.502098 , abstract =

    work page doi:10.1145/502090.502098 2001

  48. [48]

    On the power of unique 2-prover 1-round games

    Khot, Subhash , month = may, year =. On the power of unique 2-prover 1-round games , isbn =. doi:10.1145/509907.510017 , booktitle =

    work page doi:10.1145/509907.510017

  49. [49]

    Steinberg, Brooke , month = aug, year =. North

  50. [50]

    Balancing

    Verma, Shresth and Boehmer, Niclas and Kong, Lingkai and Tambe, Milind , month = sep, year =. Balancing. doi:10.48550/arXiv.2408.12112 , abstract =

    work page doi:10.48550/arxiv.2408.12112

  51. [51]

    Financial Times , author =

    Algorithms are deciding who gets organ transplants. Financial Times , author =

  52. [52]

    Frontiers in Artificial Intelligence , author =

    Smart match: revolutionizing organ allocation through artificial intelligence , volume =. Frontiers in Artificial Intelligence , author =. 2024 , pmid =. doi:10.3389/frai.2024.1364149 , abstract =

    work page doi:10.3389/frai.2024.1364149 2024

  53. [53]

    Group fairness for the allocation of indivisible goods , isbn =

    Conitzer, Vincent and Freeman, Rupert and Shah, Nisarg and Vaughan, Jennifer Wormian , month = jan, year =. Group fairness for the allocation of indivisible goods , isbn =. Thirty-. doi:10.1609/aaai.v33i01.33011853 , abstract =

    work page doi:10.1609/aaai.v33i01.33011853

  54. [54]

    IEEE Transactions on Parallel and Distributed Systems , author =

    Multi-. IEEE Transactions on Parallel and Distributed Systems , author =. 2015 , keywords =. doi:10.1109/TPDS.2014.2362139 , abstract =

    work page doi:10.1109/tpds.2014.2362139 2015

  55. [55]

    SIAM Journal on Applied Mathematics , author =

    Bounds on. SIAM Journal on Applied Mathematics , author =. 1969 , pages =

    1969

  56. [56]

    American Journal of Human Biology , author =

    Food deserts and nutritional risk in. American Journal of Human Biology , author =. 2012 , pages =. doi:10.1002/ajhb.22270 , abstract =

    work page doi:10.1002/ajhb.22270 2012

  57. [57]

    Chouldechova, Alexandra , month = jun, year =. Fair. doi:10.1089/big.2016.0047 , abstract =

    work page doi:10.1089/big.2016.0047 2016

  58. [58]

    Symposium on Discrete Algorithms (SODA) , author =

    Algorithms for facility location problems with outliers , issn =. Symposium on Discrete Algorithms (SODA) , author =. 2001 , keywords =

    2001

  59. [59]

    SIAM Journal on Computing , author =

    Worst-. SIAM Journal on Computing , author =. 1975 , pages =. doi:10.1137/0204021 , abstract =

    work page doi:10.1137/0204021 1975

  60. [60]

    and Shmoys, David B

    Williamson, David P. and Shmoys, David B. , year =. The

  61. [61]

    , month = apr, year =

    Johnson, David S. , month = apr, year =. Approximation algorithms for combinatorial problems , url =. Symposium on

  62. [62]

    Mathematical Programming , author =

    An improved approximation ratio for the minimum latency problem , volume =. Mathematical Programming , author =. 1998 , keywords =

    1998

  63. [63]

    and Lenstra, Jan Karel and Rinnooy Kan, Alexander H

    Lawler, E.L. and Lenstra, Jan Karel and Rinnooy Kan, Alexander H. G. and Shmoys, David B. , year =. The

  64. [64]

    Hardy, G. H. and Littlewood, J. E. and Pólya, George , year =. Inequalities , url =

  65. [65]

    Boyd, Stephen and Vandenberghe, Lieven , year =. Convex

  66. [66]

    Mathematical Programming , author =

    An approximation algorithm for the generalized assignment problem , volume =. Mathematical Programming , author =. 1993 , keywords =

    1993

  67. [67]

    Algorithmica , author =

    Approximating. Algorithmica , author =. 2004 , keywords =

    2004

  68. [68]

    The Bell System Technical Journal , author =

    Bounds for certain multiprocessing anomalies , volume =. The Bell System Technical Journal , author =. 1966 , pages =. doi:10.1002/j.1538-7305.1966.tb01709.x , abstract =

    work page doi:10.1002/j.1538-7305.1966.tb01709.x 1966

  69. [69]

    Journal of the ACM , author =

    A threshold of ln n for approximating set cover , volume =. Journal of the ACM , author =. 1998 , keywords =

    1998

  70. [70]

    The minimum latency problem , isbn =

    Blum, Avrim and Chalasani, Prasad and Coppersmith, Don and Pulleyblank, Bill and Raghavan, Prabhakar and Sudan, Madhu , year =. The minimum latency problem , isbn =. Symposium on. doi:10.1145/195058.195125 , language =

    work page doi:10.1145/195058.195125

  71. [71]

    Farhadi, Majid and Toriello, Alejandro and Tetali, Prasad , year =. The. Applied and. doi:10.1137/1.9781611976830.19 , pages =

    work page doi:10.1137/1.9781611976830.19

  72. [72]

    Manshadi, Vahideh and Rodilitz, Scott , month = jul, year =. Online. Economics and

  73. [73]

    and Dimitrijević, Bojan R

    Nikolić, Goran S. and Dimitrijević, Bojan R. and Nikolić, Tatjana R. and Stojcev, Mile K. , month = jun, year =. A. doi:10.1109/ICEST55168.2022.9828625 , abstract =

    work page doi:10.1109/icest55168.2022.9828625 2022

  74. [74]

    Interpolating between k-

    Chakrabarty, Deeparnab and Swamy, Chaitanya , year =. Interpolating between k-. International

  75. [75]

    Azar, Yossi and Taub, Shai , year =. All-. Algorithm. doi:10.1007/978-3-540-27810-8_26 , abstract =

    work page doi:10.1007/978-3-540-27810-8_26

  76. [76]

    Fairness measures for resource allocation , doi =

    Kumar, Amit and Kleinberg, Jon , month = nov, year =. Fairness measures for resource allocation , doi =. Symposium on

  77. [77]

    Unconstrained submodular maximization with constant adaptive complexity , booktitle =

    Chakrabarty, Deeparnab and Swamy, Chaitanya , month = jun, year =. Approximation algorithms for minimum norm and ordered optimization problems , isbn =. Symposium on. doi:10.1145/3313276.3316322 , abstract =

    work page doi:10.1145/3313276.3316322

  78. [78]

    Gupta, Swati and Moondra, Jai and Singh, Mohit , month = jul, year =. Which. Economics and

  79. [79]

    Inherent

    Kleinberg, Jon , month = jun, year =. Inherent. 2018. doi:10.1145/3219617.3219634 , abstract =

    work page doi:10.1145/3219617.3219634 2018

  80. [80]

    Equality of

    Hardt, Moritz and Price, Eric and Price, Eric and Srebro, Nati , year =. Equality of. Advances in

Showing first 80 references.