pith. sign in

arxiv: 2606.12433 · v1 · pith:VKAZI27Onew · submitted 2026-05-15 · 💻 cs.CY · cs.CL

Marginal Alignment Does Not Guarantee Joint-Distribution Fidelity: An Official-Reference Audit of Nemotron-Personas-Korea with Cross-Locale Replication

Joonhyung Bae This is my paper

Pith reviewed 2026-06-30 19:02 UTC · model grok-4.3

classification 💻 cs.CY cs.CL
keywords synthetic personasjoint distributionmarginal alignmentaudit primitiveNemotron-Personas-Koreaindependence assumptiondemographic fidelityKOSIS reference
0
0 comments X

The pith

Marginal alignment does not guarantee joint-distribution fidelity in synthetic personas

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

Synthetic persona datasets often cite alignment with official marginal demographics as a basis for trust, yet users rely on them as joint structures across attributes like age, sex, occupation and education. The paper shows that marginal alignment alone does not preserve these joints. It introduces the Independence-Assumption Footprint (IAF) audit primitive that checks synthetic joints against official references for combinations the dataset card treats as independent. Applied to Nemotron-Personas-Korea, the dataset matches KOSIS marginals but fails on three joints: major-by-occupation distribution, age profile of military service, and female representation in male-dominated occupations. The audit is replicated across six further locales and finds locale-dependent rather than universal results.

Core claim

Marginal alignment with official statistics does not imply fidelity in the joint distributions of synthetic persona attributes; the IAF audit on Nemotron-Personas-Korea confirms mismatches in three specific joints despite marginal agreement, with the strict screening verdict mapping-dependent and age-robust under direct standardisation.

What carries the argument

Independence-Assumption Footprint (IAF): an audit primitive that compares synthetic attribute joints to external official or institutional references for combinations the dataset card documents as treated independently.

Load-bearing premise

The attribute combinations a dataset card itself documents as treated independently are the appropriate ones to audit, and external official references provide accurate and directly comparable joint distributions.

What would settle it

A direct comparison in which the synthetic major-by-occupation distribution exactly matches the KEIS graduate-universe joint table would falsify the mismatch claim for that combination.

read the original abstract

Synthetic persona datasets cite alignment with official demographics as a basis for trust, yet downstream users consume them as joint structures across age, sex, region, occupation, education, name, and institutional status. Marginal alignment does not imply that these joints are preserved. We propose the Independence-Assumption Footprint (IAF), an audit primitive that operates on the attribute combinations a dataset card itself documents as treated independently. For each such combination, IAF compares the synthetic joint against an external official or institutional reference, using direct joint tables where available and rule-implied checks otherwise. Applied to NVIDIA Nemotron-Personas-Korea (one million Korean synthetic personas), IAF finds that NPK aligns with KOSIS marginals while three joints fail. The major-by-occupation distribution against the KEIS graduate universe carries a large conditional mismatch. The age profile of military service is institutionally inconsistent. Female representation in male-dominated occupations is substantially over-flattened toward parity, with the strict screening verdict mapping-dependent and age-robust under direct standardisation. A transferability demonstration across six further NPK locales finds locale-dependent rather than universal diagnostics, with reference-taxonomy cardinality confounding cross-locale flag counts. For synthetic personas used as silicon samples, marginal claims must therefore be paired with disclosure-anchored joint audits before reuse. The released audit artefacts (reference manifests, occupational crosswalks, derived metrics, reproducibility scripts) instantiate this protocol on the NPK family and are released for retargeting at other synthetic persona resources.

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

2 major / 2 minor

Summary. The paper claims that marginal alignment of synthetic persona datasets with official statistics does not guarantee fidelity of the underlying joint distributions. It introduces the Independence-Assumption Footprint (IAF) audit primitive, which targets attribute combinations documented as independent in a dataset card and compares the synthetic joints to external references (KOSIS, KEIS) via direct tables or rule-implied checks. Applied to Nemotron-Personas-Korea, IAF reports alignment on marginals but failures on three joints (major-by-occupation, military-service age profile, and female representation in male-dominated occupations); a cross-locale replication across six further NPK locales shows locale-dependent rather than universal diagnostics, with some results flagged as mapping-dependent. The manuscript releases reference manifests, occupational crosswalks, derived metrics, and reproducibility scripts.

Significance. If the reported joint mismatches are robust to taxonomy and definition differences, the result underscores a practical limitation in current validation practices for synthetic data used as silicon samples. The explicit release of crosswalks, manifests, and scripts is a strength that supports independent verification and retargeting of the protocol, distinguishing the work from purely descriptive audits.

major comments (2)
  1. [Abstract] Abstract (NPK application paragraph): The central claim that three specific joints fail rests on the premise that KOSIS/KEIS joint tables are definitionally aligned with NPK attribute encodings (occupation codes, military-service age windows, gender-occupation mappings). The text acknowledges 'reference-taxonomy cardinality confounding' and that one verdict is 'mapping-dependent,' but does not discharge this by reporting explicit crosswalk verification or sensitivity to alternative binning choices; without that, the mismatches could be artifacts of non-comparable categories rather than true distribution failures.
  2. [Abstract] Abstract (IAF definition and application): The IAF is described as using 'direct joint tables where available and rule-implied checks otherwise,' yet the manuscript provides no explicit statement of the decision rules, statistical metric (e.g., conditional KL, total variation), or failure threshold applied to declare a joint mismatch. This leaves the quantitative support for the three failures only partially specified in the main text, even though artefacts are released.
minor comments (2)
  1. [Abstract] The abstract uses the term 'strict screening verdict' without prior definition; a brief parenthetical gloss on first use would improve readability.
  2. [Cross-locale replication] Cross-locale section: the statement that diagnostics are 'locale-dependent rather than universal' would be strengthened by a small table enumerating which joints were audited per locale and which flags were raised.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on the abstract. We address each point below and will incorporate clarifications to improve self-containment.

read point-by-point responses

  1. Referee: [Abstract] Abstract (NPK application paragraph): The central claim that three specific joints fail rests on the premise that KOSIS/KEIS joint tables are definitionally aligned with NPK attribute encodings (occupation codes, military-service age windows, gender-occupation mappings). The text acknowledges 'reference-taxonomy cardinality confounding' and that one verdict is 'mapping-dependent,' but does not discharge this by reporting explicit crosswalk verification or sensitivity to alternative binning choices; without that, the mismatches could be artifacts of non-comparable categories rather than true distribution failures.

    Authors: The manuscript releases occupational crosswalks, reference manifests, and reproducibility scripts specifically to support verification of definitional alignment between NPK encodings and KOSIS/KEIS tables. The full text describes crosswalk construction and explicitly flags the mapping-dependent verdict for one joint. While the abstract summarizes these releases, we agree an explicit reference to the crosswalks would strengthen the claim. We will revise the abstract to note the released crosswalks and that sensitivity checks on binning are provided in the artefacts. This addresses the concern without altering the reported mismatches. revision: yes

  2. Referee: [Abstract] Abstract (IAF definition and application): The IAF is described as using 'direct joint tables where available and rule-implied checks otherwise,' yet the manuscript provides no explicit statement of the decision rules, statistical metric (e.g., conditional KL, total variation), or failure threshold applied to declare a joint mismatch. This leaves the quantitative support for the three failures only partially specified in the main text, even though artefacts are released.

    Authors: The full manuscript defines the IAF decision rules, metrics (conditional mismatch against official references), and failure thresholds in the methods section, with exact values in the released artefacts. The abstract condenses this for brevity. To make the abstract self-contained, we will add a concise clause on the failure criterion. This is a clarification revision. revision: yes

Circularity Check

0 steps flagged

No circularity: audit compares synthetic joints to independent external references

full rationale

The paper defines the IAF primitive to audit attribute combinations documented as independent in the dataset card, then compares the resulting synthetic joints directly against external official sources (KOSIS marginals, KEIS graduate tables, institutional military rules). The three reported failures are empirical mismatches against these outside references, not quantities derived from parameters fitted inside the paper or reduced by any equation to the inputs. No self-citations, uniqueness theorems, or ansatzes are invoked as load-bearing steps. The taxonomy-comparability concern raised in the skeptic note is a potential validity or measurement issue, not a circular reduction of the claimed result to its own construction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The central claim rests on the premise that official statistics constitute reliable joint-distribution benchmarks and that the synthetic dataset's documented independence assumptions correctly identify the combinations that require auditing.

axioms (2)
  • domain assumption Official statistics from KOSIS and KEIS accurately represent the true joint distributions of the referenced attributes.
    The audit treats these sources as ground truth for mismatch detection without additional validation steps described in the abstract.
  • domain assumption The combinations documented in the dataset card as treated independently are the relevant ones for fidelity assessment.
    This premise defines the scope of the IAF primitive.
invented entities (1)
  • Independence-Assumption Footprint (IAF) no independent evidence
    purpose: Audit primitive that flags joint-distribution deviations based on dataset-card independence assumptions.
    New method introduced by the paper to perform the reported checks.

pith-pipeline@v0.9.1-grok · 5810 in / 1510 out tokens · 41905 ms · 2026-06-30T19:02:08.230930+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

90 extracted references · 20 canonical work pages · 6 internal anchors

  1. [1]

    IEEE Journal on Emerging and Selected Topics in Circuits and Systems , volume =

    Auditing and Generating Synthetic Data with Controllable Trust Trade-offs , author =. IEEE Journal on Emerging and Selected Topics in Circuits and Systems , volume =. 2024 , doi =

    2024

  2. [2]

    Communications of the ACM , volume =

    Datasheets for Datasets , author =. Communications of the ACM , volume =. 2021 , doi =

    2021

  3. [3]

    2025 , url =

    Banyas, Peter and Sharma, Shristi and Simmons, Alistair and Vispute, Atharva , journal =. 2025 , url =

    2025

  4. [4]

    2025 , url =

    Li, Ang and Chen, Haozhe and Namkoong, Hongseok and Peng, Tianyi , journal =. 2025 , url =

    2025

  5. [5]

    Whose Personae? Synthetic Persona Experiments in

    Batzner, Jan and Stocker, Volker and Tang, Bingjun and Natarajan, Anusha and Chen, Qinhao and Schmid, Stefan and Kasneci, Gjergji , journal =. Whose Personae? Synthetic Persona Experiments in. 2025 , url =

    2025

  6. [6]

    The Personality Trap: How

    Amidei, Jacopo and Ferreira, Gregorio and Mu. The Personality Trap: How. arXiv preprint arXiv:2602.03334 , year =

    work page arXiv

  7. [7]

    arXiv preprint arXiv:2512.12775 , year =

    Persistent Personas? Role-Playing, Instruction Following, and Safety in Extended Interactions , author =. arXiv preprint arXiv:2512.12775 , year =

    work page arXiv

  8. [8]

    Cao, Hongliu and Thomas, Eoin and Acuna Agost, Rodrigo , journal =. When. 2026 , url =

    2026

  9. [9]

    Population-Aligned Persona Generation for

    Hu, Zhengyu and Lian, Jianxun and Xiao, Zheyuan and Xiong, Max and Lei, Yuxuan and Wang, Tianfu and Ding, Kaize and Xiao, Ziang and Yuan, Nicholas Jing and Xie, Xing , journal =. Population-Aligned Persona Generation for. 2025 , url =

    2025

  10. [10]

    2026 , month = apr, howpublished =

    Nemotron-Personas-Korea: Synthetic Personas Aligned to Real-World Distributions for Korea , author =. 2026 , month = apr, howpublished =

    2026

  11. [11]

    2026 , howpublished =

    Nemotron-Personas Collection , author =. 2026 , howpublished =

    2026

  12. [12]

    2025 , howpublished =

    Nemotron-Personas-USA , author =. 2025 , howpublished =

    2025

  13. [13]

    2025 , howpublished =

    Nemotron-Personas-Japan , author =. 2025 , howpublished =

    2025

  14. [14]

    2025 , howpublished =

    Nemotron-Personas-India , author =. 2025 , howpublished =

    2025

  15. [15]

    2026 , howpublished =

    Nemotron-Personas-Brazil: Co-Designed Data for Sovereign AI , author =. 2026 , howpublished =

    2026

  16. [16]

    2026 , howpublished =

    Nemotron-Personas-Singapore: Co-Designed Data for Sovereign AI , author =. 2026 , howpublished =

    2026

  17. [17]

    2026 , howpublished =

    Nemotron-Personas-France , author =. 2026 , howpublished =

    2026

  18. [18]

    2026 , howpublished =

    KOSIS OpenAPI Developer Guide , author =. 2026 , howpublished =

    2026

  19. [19]

    2024 , howpublished =

    Annual Report: Periodic Labour Force Survey (. 2024 , howpublished =

    2024

  20. [20]

    2018 , howpublished =

    Pesquisa Nacional por Amostra de Domic. 2018 , howpublished =

    2018

  21. [21]

    2025 , howpublished =

    Labour Force in Singapore 2024 , author =. 2025 , howpublished =

    2024

  22. [22]

    2026 , howpublished =

    Cat. 2026 , howpublished =

    2026

  23. [23]

    2025 , howpublished =

    American Community Survey 2024 1-Year Public Use Microdata Sample , author =. 2025 , howpublished =

    2024

  24. [24]

    2020 , howpublished =

    2020 Population Census: Summary of the Results and Statistical Tables , author =. 2020 , howpublished =

    2020

  25. [25]

    2011 , howpublished =

    Census of India 2011: Data Tables , author =. 2011 , howpublished =

    2011

  26. [26]

    2022 , howpublished =

    Censo Demogr. 2022 , howpublished =

    2022

  27. [27]

    2021 , howpublished =

    Census of Population 2020: Statistical Release 2 (. 2021 , howpublished =

    2020

  28. [28]

    2022 , howpublished =

    Recensement de la population: Fichiers d. 2022 , howpublished =

    2022

  29. [29]

    2014 , month = may, howpublished =

    Family-Relation Statistics Opening Press Release , author =. 2014 , month = may, howpublished =

    2014

  30. [30]

    2026 , howpublished =

    Electronic Family Registration System Statistics Portal , author =. 2026 , howpublished =

    2026

  31. [31]

    2024 , eprint =

    Large Language Models are Inconsistent and Biased Evaluators , author =. 2024 , eprint =

    2024

  32. [32]

    and Zhang, Hao and Gonzalez, Joseph E

    Zheng, Lianmin and Chiang, Wei-Lin and Sheng, Ying and Zhuang, Siyuan and Wu, Zhanghao and Zhuang, Yonghao and Lin, Zi and Li, Zhuohan and Li, Dacheng and Xing, Eric P. and Zhang, Hao and Gonzalez, Joseph E. and Stoica, Ion , booktitle =. Judging. 2023 , eprint =

    2023

  33. [33]

    and Feng, Shi , booktitle =

    Panickssery, Arjun and Bowman, Samuel R. and Feng, Shi , booktitle =. 2024 , eprint =

    2024

  34. [34]

    Large Language Models are not Fair Evaluators

    Large Language Models are not Fair Evaluators , author =. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages =. 2024 , doi =. 2305.17926 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  35. [35]

    Political Analysis , volume =

    Out of One, Many: Using Language Models to Simulate Human Samples , author =. Political Analysis , volume =. 2023 , doi =. 2209.06899 , archivePrefix =

    work page arXiv 2023

  36. [36]

    Proceedings of the 40th International Conference on Machine Learning , series =

    Whose Opinions Do Language Models Reflect? , author =. Proceedings of the 40th International Conference on Machine Learning , series =. 2023 , eprint =

    2023

  37. [37]

    2023 , doi =

    Cheng, Myra and Piccardi, Tiziano and Yang, Diyi , booktitle =. 2023 , doi =. 2310.11501 , archivePrefix =

    work page arXiv 2023

  38. [38]

    2024 , doi =

    Jin, Jiho and Kim, Jiseon and Lee, Nayeon and Yoo, Haneul and Oh, Alice and Lee, Hwaran , journal =. 2024 , doi =. 2307.16778 , archivePrefix =

    work page arXiv 2024

  39. [39]

    Break the Checkbox: Challenging Closed-Style Evaluations of Cultural Alignment in

    Kabir, Mohsinul and Abrar, Ajwad and Ananiadou, Sophia , booktitle =. Break the Checkbox: Challenging Closed-Style Evaluations of Cultural Alignment in. 2025 , doi =. 2502.08045 , archivePrefix =

    work page arXiv 2025

  40. [40]

    2024 , address =

    Son, Guijin and Lee, Hanwool and Kim, Suwan and Kim, Huiseo and Lee, Jaecheol and Yeom, Je Won and Jung, Jihyu and Kim, Jungwoo and Kim, Songseong , booktitle =. 2024 , address =. 2309.02706 , doi =

    work page arXiv 2024

  41. [41]

    Advances in Neural Information Processing Systems (NeurIPS) , pages =

    Equality of Opportunity in Supervised Learning , author =. Advances in Neural Information Processing Systems (NeurIPS) , pages =. 2016 , eprint =

    2016

  42. [42]

    Certifying and removing disparate impact

    Certifying and Removing Disparate Impact , author =. Proceedings of the 21st ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , pages =. 2015 , doi =. 1412.3756 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  43. [43]

    Fair prediction with disparate impact: A study of bias in recidivism prediction instruments

    Fair Prediction with Disparate Impact: A Study of Bias in Recidivism Prediction Instruments , author =. Big Data , volume =. 2017 , doi =. 1610.07524 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  44. [44]

    , booktitle =

    Parrish, Alicia and Chen, Angelica and Nangia, Nikita and Padmakumar, Vishakh and Phang, Jason and Thompson, Jana and Htut, Phu Mon and Bowman, Samuel R. , booktitle =. 2022 , doi =

    2022

  45. [45]

    Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency (FAccT '22) , pages =

    Data Cards: Purposeful and Transparent Dataset Documentation for Responsible AI , author =. Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency (FAccT '22) , pages =. 2022 , doi =. 2204.01075 , archivePrefix =

    work page arXiv 2022

  46. [46]

    Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT '21) , pages =

    Towards Accountability for Machine Learning Datasets: Practices from Software Engineering and Infrastructure , author =. Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT '21) , pages =. 2021 , doi =. 2010.13561 , archivePrefix =

    work page arXiv 2021

  47. [47]

    Model Cards for Model Reporting

    Model Cards for Model Reporting , author =. Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT* '19) , pages =. 2019 , doi =. 1810.03993 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  48. [48]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL), Volume 1: Long Papers , pages =

    Subtle Biases Need Subtler Measures: Dual Metrics for Evaluating Representative and Affinity Bias in Large Language Models , author =. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL), Volume 1: Long Papers , pages =. 2024 , doi =

    2024

  49. [49]

    Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =

    Biased Tales: Cultural and Topic Bias in Generating Children's Stories , author =. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =. 2025 , doi =. 2509.07908 , archivePrefix =

    work page arXiv 2025

  50. [50]

    Scaling Synthetic Data Creation with 1,000,000,000 Personas

    Ge, Tao and Chan, Xin and Wang, Xiaoyang and Yu, Dian and Mi, Haitao and Yu, Dong , year =. Scaling Synthetic Data Creation with 1. 2406.20094 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv

  51. [51]

    2023 , doi =

    Lee, Hwaran and Hong, Seokhee and Park, Joonsuk and Kim, Takyoung and Kim, Gunhee and Ha, Jung-woo , booktitle =. 2023 , doi =. 2305.17701 , archivePrefix =

    work page arXiv 2023

  52. [52]

    2020 , howpublished =

    Graduates Occupational Mobility Survey (GOMS) Public Microdata, 2019 cohort / 2020 wave , author =. 2020 , howpublished =

    2019

  53. [53]

    2024 , howpublished =

    2024 Military Manpower Administration Yearbook (2024 Korean military statistics yearbook) , author =. 2024 , howpublished =

    2024

  54. [54]

    Language Resources and Evaluation , volume =

    In no uncertain terms: a dataset for monolingual and multilingual automatic term extraction from comparable corpora , author =. Language Resources and Evaluation , volume =. 2020 , doi =

    2020

  55. [55]

    Enriching the

    Song, Jayoung and Lim, KyungTae and Park, Jungyeul , journal =. Enriching the. 2026 , doi =

    2026

  56. [56]

    Predicting lexical complexity in

    Shardlow, Matthew and Evans, Richard and Zampieri, Marcos , journal =. Predicting lexical complexity in. 2022 , doi =

    2022

  57. [57]

    2024 , howpublished =

    2024

  58. [58]

    2025 , howpublished =

    Statistical Handbook of Japan 2025 , author =. 2025 , howpublished =

    2025

  59. [59]

    The Sociological Quarterly , volume =

    The Effects of Occupational Gender Segregation across Race , author =. The Sociological Quarterly , volume =. 2003 , doi =

    2003

  60. [60]

    2009 , doi =

    Causality: Models, Reasoning, and Inference , author =. 2009 , doi =

    2009

  61. [61]

    Psychological Bulletin , volume =

    Construct validity in psychological tests , author =. Psychological Bulletin , volume =. 1955 , doi =

    1955

  62. [62]

    Transactions of the Association for Computational Linguistics , volume =

    Data Statements for Natural Language Processing: Toward Mitigating System Bias and Enabling Better Science , author =. Transactions of the Association for Computational Linguistics , volume =. 2018 , doi =

    2018

  63. [63]

    Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =

    The ``Problem'' of Human Label Variation: On Ground Truth in Data, Modeling and Evaluation , author =. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages =. 2022 , doi =

    2022

  64. [64]

    Political Analysis , volume =

    Synthetic Replacements for Human Survey Data? The Perils of Large Language Models , author =. Political Analysis , volume =. 2024 , doi =

    2024

  65. [65]

    Advances in Neural Information Processing Systems 37 (NeurIPS 2024) , year =

    Questioning the Survey Responses of Large Language Models , author =. Advances in Neural Information Processing Systems 37 (NeurIPS 2024) , year =. 2306.07951 , archivePrefix =

    work page arXiv 2024

  66. [66]

    Proceedings of the 40th International Conference on Machine Learning (ICML) , series =

    Using Large Language Models to Simulate Multiple Humans and Replicate Human Subject Studies , author =. Proceedings of the 40th International Conference on Machine Learning (ICML) , series =. 2023 , eprint =

    2023

  67. [67]

    Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology (UIST '23) , articleno =

    Generative Agents: Interactive Simulacra of Human Behavior , author =. Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology (UIST '23) , articleno =. 2023 , doi =

    2023

  68. [68]

    Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing , pages =

    Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints , author =. Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing , pages =. 2017 , doi =

    2017

  69. [69]

    Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency (FAccT '23) , pages =

    Easily Accessible Text-to-Image Generation Amplifies Demographic Stereotypes at Large Scale , author =. Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency (FAccT '23) , pages =. 2023 , doi =

    2023

  70. [70]

    2024 , doi =

    Shumailov, Ilia and Shumaylov, Zakhar and Zhao, Yiren and Papernot, Nicolas and Anderson, Ross and Gal, Yarin , journal =. 2024 , doi =

    2024

  71. [71]

    Understanding the Effects of

    Kirk, Robert and Mediratta, Ishita and Nalmpantis, Christoforos and Luketina, Jelena and Hambro, Eric and Grefenstette, Edward and Raileanu, Roberta , booktitle =. Understanding the Effects of. 2024 , eprint =

    2024

  72. [72]

    Replacing Judges with Juries: Evaluating

    Verga, Pat and Hofst. Replacing Judges with Juries: Evaluating. 2024 , eprint =

    2024

  73. [73]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages =

    Bavaresco, Anna and Bernardi, Raffaella and Bertolazzi, Leonardo and Elliott, Desmond and Fern. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages =. 2025 , doi =. 2406.18403 , archivePrefix =

    work page arXiv 2025

  74. [74]

    Proceedings of the 1st ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization (EAAMO) , articleno =

    A Framework for Understanding Sources of Harm throughout the Machine Learning Life Cycle , author =. Proceedings of the 1st ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization (EAAMO) , articleno =. 2021 , doi =

    2021

  75. [75]

    Towards Measuring the Representation of Subjective Global Opinions in Language Models

    Towards Measuring the Representation of Subjective Global Opinions in Language Models , author =. Proceedings of the First Conference on Language Modeling (COLM 2024) , year =. 2306.16388 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  76. [76]

    Sampling Techniques , author =

  77. [77]

    1975 , doi =

    Discrete Multivariate Analysis: Theory and Practice , author =. 1975 , doi =

    1975

  78. [78]

    2011 , doi =

    Synthetic Datasets for Statistical Disclosure Control: Theory and Implementation , author =. 2011 , doi =

    2011

  79. [79]

    2012 , address =

    International Standard Classification of Occupations. 2012 , address =

    2012

  80. [80]

    Journal of Survey Statistics and Methodology , volume =

    Machine Learning for Occupation Coding---A Comparison Study , author =. Journal of Survey Statistics and Methodology , volume =. 2021 , doi =

    2021

Showing first 80 references.