Pith. sign in

REVIEW 3 major objections 5 minor 28 references

Epidemiology of Model Collapse: Modeling Synthetic Data Contamination via Bilayer SIR Dynamics

T0 review · 3 major / 5 minor · reviewed 2026-07-12 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read AI data contamination behaves like an epidemic whose R0 is the geometric mean of data and model transmission rates; detection is the highest-leverage lever.

desk verdict Solid new-application paper: bilayer SIR/SIRS gives a clean geometric-mean R0 and intervention structure for ecosystem cross-contamination; GPT-2 work is only a qualitative single-chain bridge, not a test of the bilayer threshold. read the letter →

arxiv 2606.05168 v1 pith:7N4ELI7M submitted 2026-04-14 cs.CL cs.AIcs.LG

Xiangyu Wang This is my paper
classification cs.CLcs.AIcs.LG
keywords modelcollapsesyntheticdatabilayerSIRbasicreproductionnumbercross-contaminationAIecosystemdetectionfilteringSIRS
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Model collapse has been studied as a single chain of models training on their own output. This paper argues that the real problem is cross-contamination across a shared ecosystem: models train on synthetic text from other models, publish new synthetic text, and re-pollute the common corpora. It maps that loop onto a bilayer SIR/SIRS epidemic model, with one layer for data corpora and one for AI models, linked by cross-layer infection. The resulting basic reproduction number is the geometric mean of the two layers’ transmission-to-recovery ratios; when it exceeds 1 the system heads to an endemic contaminated state. Scenario calibrations from public AI-text prevalence figures put the system above that threshold, and sensitivity analysis ranks synthetic-text detection as the single most powerful parameter. GPT-2 contamination-chain experiments show dose-response degradation and diversity loss that line up with the supercritical picture, while matched-budget multi-source runs give only a modest buffer that disappears once real data is mixed back in. The practical message is that filtering contaminated data and keeping a large clean-trained fraction of models matter more than diversifying contamination sources.

What carries the argument

The bilayer basic reproduction number R0 = sqrt(β_D β_M / [(γ_D + μ_D)(γ_M + μ_M)]), obtained by the Next Generation Matrix on the infected subsystem (I_D, I_M). Its geometric-mean structure encodes that a full contamination generation must traverse both layers, so interventions that raise either recovery rate or lower either transmission rate can drive the whole system subcritical.

What would settle it

Measure synthetic-text fractions and clean-retraining rates at ecosystem scale; if the resulting R0 estimate is stably below 1 while observed model quality and diversity continue to degrade under realistic mixed training, or if raising detection coverage fails to reduce measured contamination while other parameters stay fixed, the central claim fails.

Watch

Extended reading notes

Core claim

Synthetic-data cross-contamination in the AI ecosystem can be treated as a bilayer SIR/SIRS epidemic whose basic reproduction number is R0 = sqrt(β_D β_M / [(γ_D + μ_D)(γ_M + μ_M)]). Under three illustrative calibrations drawn from public AI-text prevalence data, R0 > 1; Sobol analysis identifies data detection γ_D as the highest-leverage parameter; and GPT-2 chains exhibit dose-response quality and diversity loss qualitatively consistent with the supercritical/near-critical regime.

Load-bearing premise

That continuous contamination levels can be split into binary clean/contaminated/recovered compartments by a fixed quality threshold without changing the qualitative threshold structure or the ranking of interventions.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes a bilayer coupled SIR/SIRS mean-field model of synthetic-data contamination in the AI ecosystem, treating data corpora and models as two interacting populations with cross-layer transmission. It derives R0 = sqrt(β_D β_M / [(γ_D+μ_D)(γ_M+μ_M)]) via the Next Generation Matrix, establishes standard DFE stability, endemic existence, and transcritical bifurcation results, and recommends the SIRS variant for immunity waning. Illustrative scenario calibrations from public AI-text prevalence data yield R0 > 1; Sobol analysis ranks data detection γ_D highest-leverage. An ABM on bipartite networks checks mean-field consistency for dense graphs. GPT-2 contamination chains (192 runs) show dose-response perplexity and diversity degradation; matched-budget multi-source experiments (1,088 runs) give borderline attenuation at α=1 that vanishes at α=0.5. Intervention analysis, under model assumptions, favors detection/filtering and herd immunity.

Significance. If the framing holds, the paper supplies a usable epidemic vocabulary (R0, herd immunity, cross-layer leverage) for ecosystem-level synthetic-data contamination, going beyond single-chain model-collapse analyses. Strengths include a clean, algebraically correct NGM derivation with explicit cancellation of cross-population ratios; numerical verification of threshold/bifurcation claims on large random ensembles; transparent labeling of calibration as illustrative; an ABM consistency check with quantified breakdown under heterogeneity; and a large, matched-budget GPT-2 experimental suite with pre-specified one-sided tests. The geometric-mean R0 structure and the intervention ranking that follows from it are the main conceptual contributions. The work is phenomenological applied theory rather than a fitted ecosystem measurement, which limits policy weight but does not erase the value of the framework.

major comments (3)
  1. [Section 6 / Abstract] Section 6.4 maps α to transmission intensity and (1−α) to recovery, and the abstract/intro package the GPT-2 results as “qualitatively consistent with the threshold picture.” The experiments are single-lineage recursive fine-tunes (or fixed-pool multi-source mixes) that never instantiate two interacting populations, never measure S/I/R compartment fractions, and never estimate β or γ rates. Dose-response and Distinct-2 collapse are already predicted by single-chain collapse theory (Shumailov et al., Dohmatob et al.). The only bilayer-specific empirical claim—source diversity K via β_eff_M(K)=β_M/f(K)—is borderline at α=1 (one-sided p=0.047, ~2 PPL) and null at α=0.5. The manuscript should either (a) reframe Section 6 as an empirical bridge to recursive degradation only, not to bilayer R0 supercriticality, or (b) add an analysis that actually probes cross-layer structure (e.g., separate d
  2. [Section 4.2–4.3, Section 7, Table 2] Section 4 and Table 2 present three scenarios with R0 ∈ {1.10, 2.62, 6.63} and P(R0>1)=98.2% from Sobol sampling, then Section 7 ranks interventions that drive R0 below 1. The paper correctly labels calibration as illustrative, but the intervention conclusions (watermark+filtering and herd immunity as sole single strategies achieving R0<1; γ_D as highest leverage) inherit the assumed ranges and the algebraic form of R0 rather than measured elasticities. The load-bearing claim for readers is that detection is the practical lever; this needs a sharper separation between “model-conditional ranking under assumed ranges” and any claim about the real ecosystem. A short sensitivity table showing how the ranking changes under alternative prior ranges (especially γ_D coverage and β_D growth) would make the conditionality operational.
  3. [Section 3.1] Section 3.1 operationalizes continuous contamination via a threshold τ into binary S/I/R states, noting that R0 is independent of τ. That algebraic independence is correct, but all empirical mapping (α as contamination fraction) and intervention interpretation rest on the partition remaining meaningful for quality degradation. The manuscript does not show that intervention rankings or endemic levels are robust to alternative τ choices or to a continuous-state formulation. A brief continuous-state or multi-compartment sensitivity (or an explicit statement that intervention rankings are conditional on the binary partition) is needed before the herd-immunity and filtering recommendations can be treated as more than structural consequences of the ODE.
minor comments (5)
  1. [Section 3.4, Appendix I] Theorem/Proposition numbering is inconsistent in the main text (Theorem 3 vs. Proposition 3 for endemic equilibrium; Appendix I maps code names differently). Unify labels between main text, appendix, and verification code.
  2. [Figure 5] Figure 5 right panel labels pairwise comparisons “n.s.” while the text reports one-sided p=0.047 for K=1 vs K=5. Align figure annotations with the pre-specified one-sided tests and report effect sizes on the figure.
  3. [Introduction / Section E] The 74% AI-text prevalence figure is flagged as projected (Section E); the abstract and introduction still lead with it. Soften the lead sentence or move the projection caveat earlier.
  4. [Section 3.3, Appendix A.1] Eq. (7)–(8) and the implementation note correctly state that cross-population ratios cancel; a one-line remark in the main text that the code uses the simplified F would help reproducibility readers.
  5. [Table 3] Table 3 Shakespeare rows omit growth-rate and AIC columns without a clear reason in the caption; either compute them or state why they are omitted.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: R0 is standard NGM spectral radius on the bilayer infected subsystem (cross-ratios cancel algebraically), calibration is openly illustrative/scenario-based from external prevalence data, and GPT-2 runs are independent qualitative checks that do not feed parameters back into the derivation.

full rationale

The load-bearing mathematical claim is Theorem 1 / Eq. (8): R0 = sqrt(βD βM / [(γD+μD)(γM+μM)]), obtained as ρ(FV^{-1}) for the linearized infected subsystem (ID, IM). The off-diagonal product of F V^{-1} cancels the cross-population ratios exactly, yielding the geometric mean; this is ordinary Next-Generation-Matrix algebra applied to the six ODEs and does not depend on any fitted value, prevalence number, or experimental outcome. Theorems 2–4 (DFE stability, endemic existence, transcritical bifurcation) are likewise standard epidemic results verified numerically on random parameter draws, not on data. Calibration (Section 4) is explicitly labeled “illustrative scenario-based” and uses external public AI-text prevalence points only to set three example (β,γ) triples; those triples are never re-inserted into the R0 derivation or used to “predict” the same prevalence. Sobol indices are pure algebraic sensitivity of the closed-form R0. The GPT-2 chains (192 + 1 088 runs) are presented only as a “qualitative bridge” / “phenomenological analogy” that never measures compartment fractions, never estimates β or γ, and never fits the ODE; dose-response and Distinct-2 collapse are therefore independent empirical observations, not circular predictions. The sole auxiliary construct β_eff_M(K)=βM/f(K) is openly labeled a modeling hypothesis motivated by heterogeneous mixing and is tested (not assumed) by the matched-budget ablation; its weak, non-monotonic result is reported as such. No self-citations exist (single-author paper), no uniqueness theorem is imported, and no known empirical pattern is merely renamed. The derivation chain is therefore self-contained against its own inputs.

Assumptions & free parameters 4 free parameters · 4 assumptions · 2 invented entities

The mathematical skeleton is standard epidemic theory; the paper’s contribution is the domain mapping and the illustrative numbers. Free parameters are the six rate constants plus the auxiliary diversity factor; axioms are homogeneous mixing, the binary threshold partition, and the cross-layer transmission interpretation; the invented entities are the two interacting S/I/R populations themselves.

free parameters (4)
  • β_D (data contamination rate) = 0.216 month^-1 (baseline)
    Point estimate 0.216 obtained by log-linear regression on six public AI-text prevalence points plus recovery/turnover offsets; CI [0.201,0.231] used for scenarios.
  • γ_D (detection/removal rate) = 0.099 month^-1
    Order-of-magnitude product of ~85% recall and ~12% platform coverage; not measured at ecosystem scale.
  • β_M, γ_M, μ_D, μ_M, Λ_D, Λ_M = β_M=0.340, γ_M=0.060, μ_D=0.02, μ_M=0.03, Λ_D=5, Λ_M=3
    Hand-estimated from training frequency, clean-retraining fraction, obsolescence and deployment rates; used for all numerical R0 and intervention sweeps.
  • f(K) diversity attenuation factor = unspecified functional form (e.g. 1+c log K)
    Auxiliary postulate β_eff_M(K)=β_M/f(K) with f(1)=1 and f increasing; not derived from the base ODEs.
assumptions (4)
  • domain assumption Homogeneous mixing within each layer and mass-action cross-layer incidence
    Standard mean-field SIR premise; ABM section shows it fails under sparse or superspreader networks (Section 5).
  • ad hoc to paper Continuous contamination can be thresholded into binary S/I/R states without altering qualitative R0 structure
    Explicitly introduced in Section 3.1 operational definitions; τ never appears in R0 yet all empirical claims rest on the partition.
  • standard math Next-generation-matrix spectral radius correctly gives the invasion threshold for the bilayer system
    Invokes van den Driessche & Watmough (2002) and Diekmann et al.; verified numerically to machine precision.
  • standard math Waning-immunity rate δ leaves the infected-subsystem linearization (hence R0) unchanged
    Standard SIRS fact used to claim all threshold theorems carry over (Section 3.5).
invented entities (2)
  • Bilayer data-corpora / AI-models S/I/R populations with cross-layer transmission
    purpose: To turn ecosystem-level synthetic-data contamination into a two-population epidemic system whose R0 and interventions can be analyzed with existing theory.
    No independent measurement of compartment fractions exists; the mapping is purely phenomenological.
  • Source-diversity attenuation factor f(K)
    purpose: To encode the hypothesis that multi-model contamination pools reduce effective β_M.
    Postulated in Section 3.5; only weakly supported by the matched-budget experiment at α=1 and unsupported at α=0.5.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Epidemiology of Model Collapse: Modeling Synthetic Data Contamination via Bilayer SIR Dynamics." pith.science (2026). https://pith.science/paper/7N4ELI7M

@misc{pith2026260605168,
  author       = {Pith},
  title        = {Pith review of: Epidemiology of Model Collapse: Modeling Synthetic Data Contamination via Bilayer SIR Dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7N4ELI7M}},
  note         = {Machine review of arXiv:2606.05168}
}
abstract

Training on synthetic data causes model collapse, but existing analyses treat this as single-chain degradation. In reality, the AI ecosystem involves cross-contamination: models ingest synthetic data from other models, produce new synthetic text, and contaminate shared corpora. We propose a bilayer coupled SIR/SIRS framework -- a phenomenological mean-field model treating data corpora and AI models as two interacting populations, each with susceptible, infected, and recovered compartments linked by cross-layer transmission. The SIRS variant (our primary recommendation) incorporates immunity waning, reflecting that filtered corpora and retrained models remain susceptible to re-contamination. We derive the basic reproduction number $R_0 = \sqrt{\beta_D \beta_M / [(\gamma_D+\mu_D)(\gamma_M+\mu_M)]}$ via the Next Generation Matrix and apply standard epidemic threshold results to the bilayer system. Illustrative scenario-based calibration from public AI text prevalence data yields supercritical dynamics ($R_0 > 1$) across three scenarios; Sobol sensitivity analysis identifies synthetic-text detection as the highest-leverage parameter. A bipartite-network agent-based model confirms mean-field consistency ($R^2 > 0.96$) for dense networks but degrades under heterogeneity. GPT-2 contamination chain experiments (192 runs across WikiText and Shakespeare) show dose-response degradation and diversity loss qualitatively consistent with the threshold picture. Matched-budget source-diversity experiments (1,088 runs) provide suggestive evidence that multi-source mixing modestly attenuates collapse, but the effect vanishes at lower contamination fractions. Intervention analysis identifies detection-based filtering and herd immunity as the highest-leverage strategies.

Figures

Figures reproduced from arXiv: 2606.05168 by the authors.

Figure 1
Figure 1. Bilayer SIR schematic. Data corpora (top, blue) and AI models (bottom, orange) form [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. ODE trajectory under baseline parameters ( [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. R0 as a function of data infection rate (βD) and model infection rate (βM), with other parameters at baseline values. The thick contour marks R0 = 1. All three calibration scenarios (markers) lie in the supercritical region, though the optimistic scenario is near the boundary. Uncertainty propagation. The Sobol sensitivity analysis (Section 4.3) uses Saltelli’s quasi-random sampling scheme with N = 512 base samples … view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: ODE (solid) vs. ABM ensemble mean (dashed, [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Matched-budget source-diversity experiment ( [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: SIRS oscillatory dynamics for a representative configuration ( [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Sobol total-order sensitivity indices for [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: ABM threshold verification: sub-critical vs. super-critical classification across 20 parameter [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]
Figure 9
Figure 9. Figure 9: Perplexity over 8 generations for WikiText contamination chains at five contamination [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: Cross-domain comparison of contamination chain dynamics. WikiText (left) and Shake [PITH_FULL_IMAGE:figures/full_fig_p021_10.png]
Figure 11
Figure 11. Figure 11: Distinct-2 diversity across generations for [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: Training loss convergence diagnostics across generations and contamination fractions, [PITH_FULL_IMAGE:figures/full_fig_p022_12.png]
Figure 13
Figure 13. Figure 13: R0 as a function of intervention intensity for all six strategies. Only watermark-based filtering and herd immunity cross the R0 = 1 threshold. 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Relative Cost 0.0 0.5 1.0 1.5 2.0 R0 R e d u ctio n Watermark + Filtering Output Restric…
Figure 14
Figure 14. Figure 14: Pareto frontier: illustrative cost vs. R0 reduction across single-strategy sweeps (6 strategies × 20 intensity levels = 120 points). Only watermark-based filtering and herd immunity achieve R0 < 1 alone. Combined interventions (15 pairs × 9 intensity combinations = 13…
Figure 15
Figure 15. Figure 15: ODE infection trajectories under the three calibration scenarios. The pessimistic scenario [PITH_FULL_IMAGE:figures/full_fig_p024_15.png]
Figure 16
Figure 16. Figure 16: Bifurcation diagram showing endemic equilibrium infection level as a function of [PITH_FULL_IMAGE:figures/full_fig_p024_16.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

28 extracted references · 3 linked inside Pith

  1. [1]

    Self-Consuming Generative Models Go MAD

    Sina Alemohammad, Josue Casco-Rodriguez, Lorenzo Luzi, Ahmed Imtiaz Humayun, Hossein Babaei, Daniel LeJeune, and Richard G Baraniuk. Self-Consuming Generative Models Go MAD. InProceedings of the International Conference on Learning Representations (ICLR), 2024

    2024

  2. [2]

    Dynamical Models of Tuberculosis and Their Applications.Mathematical Biosciences and Engineering, volume 1, pp

    Carlos Castillo-Chavez and Baojun Song. Dynamical Models of Tuberculosis and Their Applications.Mathematical Biosciences and Engineering, volume 1, pp. 361–404, 2004. 11

    2004

  3. [3]

    Odo Diekmann, J A P Heesterbeek, and Johan A J Metz. On the Definition and the Computation of the Basic Reproduction Ratio R0 in Models for Infectious Diseases in Heterogeneous Populations.Journal of Mathematical Biology, volume 28, pp. 365–382, 1990

    1990

  4. [4]

    Strong Model Collapse

    Elvis Dohmatob, Yunzhen Feng, Arjun Subramonian, and Horia Mania. Strong Model Collapse. InProceedings of the International Conference on Learning Representations (ICLR), 2025

    2025

  5. [5]

    Is Model Collapse Inevitable? Breaking the Curse of Recursion by Accumulating Real and Synthetic Data.arXiv preprint arXiv:2404.01413, 2024

    Matthias Gerstgrasser, Rylan Schaeffer, Apratim Dey, Rafael Rafailov, Henry Sleight, John Hughes, Tomasz Korbak, Rajashree Agrawal, Dhruv Pai, Andrey Gromov, et al. Is Model Collapse Inevitable? Breaking the Curse of Recursion by Accumulating Real and Synthetic Data.arXiv preprint arXiv:2404.01413, 2024

    arXiv 2024

  6. [6]

    The Mathematics of Infectious Diseases.SIAM Review, volume 42, pp

    Herbert W Hethcote. The Mathematics of Infectious Diseases.SIAM Review, volume 42, pp. 599–653, 2000

    2000

  7. [7]

    Epidemiologi- cal Modeling of News and Rumors on Twitter.Proceedings of the Workshop on Social Network Mining and Analysis, pp

    Fang Jin, Edward Dougherty, Parang Saraf, Yang Cao, and Naren Ramakrishnan. Epidemiologi- cal Modeling of News and Rumors on Twitter.Proceedings of the Workshop on Social Network Mining and Analysis, pp. 1–9, 2013

    2013

  8. [8]

    Measuring and Modeling Computer Virus Prevalence

    Jeffrey O Kephart and Steve R White. Measuring and Modeling Computer Virus Prevalence. Proceedings of the IEEE Symposium on Security and Privacy, pp. 2–15, 1993

    1993

Show all 28 references
  1. [9]

    A Contribution to the Mathematical Theory of Epidemics.Proceedings of the Royal Society of London

    William Ogilvy Kermack and Anderson G McKendrick. A Contribution to the Mathematical Theory of Epidemics.Proceedings of the Royal Society of London. Series A, volume 115, pp. 700–721, 1927

    1927

  2. [10]

    A Watermark for Large Language Models

    John Kirchenbauer, Jonas Geiping, Yuxin Wen, Jonathan Katz, Ian Miers, and Tom Goldstein. A Watermark for Large Language Models. InProceedings of the International Conference on Machine Learning (ICML), 2023

    2023

  3. [11]

    Monitoring AI- Modified Content at Scale

    Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou. Monitoring AI- Modified Content at Scale. 2024

    2024

  4. [12]

    A Pretrainer’s Guide to Training Data: Measuring the Effects of Data Age, Domain Coverage, Quality, and Toxicity.arXiv preprint arXiv:2305.13169, 2024

    Shayne Longpre, Robert Mahari, Anthony Chen, Naana Obeid, Damien Chan, Andrea Madotto, Colin Raffel, and Harm de Vries. A Pretrainer’s Guide to Training Data: Measuring the Effects of Data Age, Domain Coverage, Quality, and Toxicity.arXiv preprint arXiv:2305.13169, 2024

    2024 arXiv

  5. [13]

    Pointer Sentinel Mixture Models.arXiv preprint arXiv:1609.07843, 2016

    Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer Sentinel Mixture Models.arXiv preprint arXiv:1609.07843, 2016

    2016 arXiv

  6. [14]

    Model Cards for Model Reporting

    Margaret Mitchell, Simone Wu, Andrew Zaldivar, Parker Barnes, Lucy Vasserman, Ben Hutchin- son, Elena Spitzer, Inioluwa Deborah Raji, and Timnit Gebru. Model Cards for Model Reporting. InProceedings of the Conference on Fairness, Accountability, and Transparency (FAT*), pp. 22...

    2019

  7. [15]

    Epidemic Processes in Complex Networks.Reviews of Modern Physics, volume 87, pp

    Romualdo Pastor-Satorras, Claudio Castellano, Piet Van Mieghem, and Alessandro Vespignani. Epidemic Processes in Complex Networks.Reviews of Modern Physics, volume 87, pp. 925–979, 2015

    2015

  8. [16]

    Lawrence Perko.Differential Equations and Dynamical Systems, Springer, 2001

    2001

  9. [17]

    Language Models are Unsupervised Multitask Learners.OpenAI Blog, 2019

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language Models are Unsupervised Multitask Learners.OpenAI Blog, 2019

    2019

  10. [18]

    Variance Based Sensitivity Analysis of Model Output

    Andrea Saltelli, Paola Annoni, Ivano Azzini, Francesca Campolongo, Marco Ratto, and Stefano Tarantola. Variance Based Sensitivity Analysis of Model Output. Design and Estimator for the Total Sensitivity Index.Computer Physics Communications, volume 181, pp. 259–270, 2010. 12

    2010

  11. [19]

    How Bad is Training on Synthetic Data? A Statistical Analysis

    Mohamed El Amine Seddik, Suei-Hai Chen, Soufiane Hayou, Pierre Youssef, and Merouane Debbah. How Bad is Training on Synthetic Data? A Statistical Analysis. InProceedings of the International Conference on Machine Learning (ICML), 2024

    2024

  12. [20]

    AI models collapse when trained on recursively generated data.Nature, volume 631, pp

    Ilia Shumailov, Zakhar Shumaylov, Yiren Zhao, Nicolas Papernot, Ross Anderson, and Yarin Gal. AI models collapse when trained on recursively generated data.Nature, volume 631, pp. 755–759, 2024

    2024

  13. [21]

    The Science of Detecting LLM-Generated Text

    Ruixiang Tang, Yu-Neng Chuang, and Xia Hu. The Science of Detecting LLM-Generated Text. Communications of the ACM, volume 67, pp. 81–90, 2024

    2024

  14. [22]

    AI-Generated Content Prevalence in Web Corpora

    Neil C Thompson and Shuning Ge. AI-Generated Content Prevalence in Web Corpora. 2024

    2024

  15. [23]

    Reproduction Numbers and Sub-threshold Endemic Equilibria for Compartmental Models of Disease Transmission.Mathematical Bio- sciences, volume 180, pp

    Pauline van den Driessche and James Watmough. Reproduction Numbers and Sub-threshold Endemic Equilibria for Compartmental Models of Disease Transmission.Mathematical Bio- sciences, volume 180, pp. 29–48, 2002

    2002

  16. [24]

    The Spread of True and False News Online

    Soroush V osoughi, Deb Roy, and Sinan Aral. The Spread of True and False News Online. Science, volume 359, pp. 1146–1151, 2018. 13 A Proof Sketches and Numerical Evidence A.1 Proof of Theorem 1: Basic Reproduction Number We apply the Next Generation Matrix (NGM) method of van ...

    2018

  17. [25]

    Infection occurs with probabilityp(Bernoulli trial)

    Infection (data nodes): For each susceptible data node, infection pressure is the traffic- weighted fraction of infected model neighbors: p=β D ·(P infected traffic)/(P all traffic). Infection occurs with probabilityp(Bernoulli trial)

  18. [26]

    3.Recovery: Each infected node recovers with probabilityγ i per step

    Infection (model nodes): For each susceptible model node, infection pressure is the fraction of infected data neighbors: p=β M ·k I /k, where kI is the number of infected data neighbors andkis the total. 3.Recovery: Each infected node recovers with probabilityγ i per step. 4.T...

  19. [27]

    Superspreaders: If enabled, a configurable fraction of model nodes have 10× traffic weight, amplifying their infection pressure on data nodes

  20. [28]

    20 realizations are run per configuration for 50 time steps (default)

    Detectors: If enabled, detector nodes scan assigned neighbors and recover each infected neighbor with probability precision×coverage per step. 20 realizations are run per configuration for 50 time steps (default). Ensemble means and standard deviations are computed at each tim...

Pith tools

Reviewed July 12, 2026 · model on record in the stance chip above.