pith. sign in

arxiv: 2510.23507 · v1 · pith:OMOC7YNKnew · submitted 2025-10-27 · 💻 cs.LG · cs.AI· cs.IT· math.IT

A Deep Latent Factor Graph Clustering with Fairness-Utility Trade-off Perspective

Siamak Ghodsi , Amjad Seyedi , Tai Le Quy , Fariba Karimi , Eirini Ntoutsi This is my paper

Pith reviewed 2026-05-21 20:21 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.ITmath.IT
keywords fair graph clusteringnonnegative tri-factorizationstatistical paritycommunity detectionfairness-utility trade-offgroup balancemodularitydeep latent factors
0
0 comments X

The pith

DFNMF uses deep nonnegative tri-factorization plus a soft parity regularizer to deliver higher group balance at comparable modularity in graph clustering.

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

The paper introduces DFNMF, an end-to-end deep nonnegative tri-factorization method for graphs that adds a soft statistical-parity regularizer to directly optimize cluster assignments. A single parameter lambda adjusts the fairness-utility balance while nonnegativity supplies parts-based factors and transparent soft memberships. The optimization relies on sparse-friendly alternating updates that scale near-linearly with the number of edges. A sympathetic reader would care because the method targets equitable partitions in applications such as community detection and resource allocation, and experiments show it often reaches higher group balance than baselines while preserving similar modularity.

Core claim

DFNMF is an end-to-end deep nonnegative tri-factorization tailored to graphs that directly optimizes cluster assignments with a soft statistical-parity regularizer. A single parameter tunes the fairness-utility balance, nonnegativity yields parts-based factors and transparent soft memberships, and across synthetic and real networks the approach achieves substantially higher group balance at comparable modularity, often dominating state-of-the-art baselines on the Pareto front.

What carries the argument

Nonnegative tri-factorization combined with a soft statistical-parity regularizer that enforces proportional group representation during modularity optimization through alternating updates.

Load-bearing premise

The combination of nonnegative tri-factorization and the soft statistical-parity regularizer produces a controllable and effective fairness-utility trade-off that generalizes beyond the tested networks without hidden biases from the alternating optimization.

What would settle it

A new network dataset in which increasing the fairness parameter lambda does not raise group balance while holding modularity steady, or in which DFNMF falls below existing baselines on the Pareto front.

Figures

Figures reproduced from arXiv: 2510.23507 by Amjad Seyedi, Eirini Ntoutsi, Fariba Karimi, Siamak Ghodsi, Tai Le Quy.

Figure 1
Figure 1. Figure 1: Fair clustering of a 16-node graph (10 Male, 6 Female) into two equal-sized clusters. Left: Utilitarian clustering yields a structure-driven partition with a 6M:2F distribution for green and 4M:4F for lavender cluster, resulting in gender imbalance. Right: Fair clustering achieves a balanced 5M:3F distribution in both clusters by swapping memberships of nodes 1 and 13. making them unsuitable for graph-stru… view at source ↗
Figure 2
Figure 2. Figure 2: DFNMF schematic and example. A 45-node graph with imbalanced gender distribution of 40%/60%(27 , 18 ) is factorized through H1, H2, H3. Two solutions illustrate the effect of λ: small λ preserves structure but yields imbalance (5:9, 5:11, 8:7); large λ improves parity (7:11, 5:7, 6:9), highlighting the utility–fairness trade-off. Algorithm 1 [Balanced] Deep Fair NMF (DFNMF) Input: The adjacency matrix of g… view at source ↗
Figure 3
Figure 3. Figure 3: SBM networks with varying node sizes: comparison of clustering and fairness metrics. Arrows ( [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Convergence curves on SBM graphs (5K and 10K [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: DFNMF hierarchy on a 60-node graph. (a) Input graph; node shapes denote sensitive groups. (b) Micro-clusters (A–L) discovered by the first layer (H1). (c) Three coarse communities obtained by aggregating micro-clusters via H1; final node–community affinities are Ψ = H1H2 [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Micro→community (H2) and node→community (Ψ = H1H2) soft memberships. scaling), shown as the green star at λ ⋆ = 100—the same setting reported in Table III. Dashed identity lines provide a balanced trade-off guide (top-right is best), and shaded curvature indicates empirical fronts. Consistent patterns emerge: DMoN and SC attain higher Q but poorer balance; fairness-oriented spectral baselines (FSC/SFSC/iFS… view at source ↗
Figure 7
Figure 7. Figure 7: Pareto plots for k=5. Blue: DFNMF across λ ∈ [10−3 , 103 ]; green star: λ ⋆ = 100 selected on the Pareto front via the ideal-point rule. Dashed identity lines mark balanced trade-offs; shaded curves indicate empirical fronts [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Selected λ ⋆ vs. k (log-scale on y). (60 → 100 → 150); DrugNet increases then plateaus (0.01 → 0.10 → 0.50); LastFM is stable at small k and rises at k=8 (0.005 → 0.005 → 0.05). (2) Modularity often peaks at a moderate k: DrugNet and Facebook achieve their best Q at k=5 (0.591 and 0.503, respectively) and then drop at k=8, while LastFM (sparser, more homophilous) shows a steady decline (0.516 → 0.420 → 0.3… view at source ↗
read the original abstract

Fair graph clustering seeks partitions that respect network structure while maintaining proportional representation across sensitive groups, with applications spanning community detection, team formation, resource allocation, and social network analysis. Many existing approaches enforce rigid constraints or rely on multi-stage pipelines (e.g., spectral embedding followed by $k$-means), limiting trade-off control, interpretability, and scalability. We introduce \emph{DFNMF}, an end-to-end deep nonnegative tri-factorization tailored to graphs that directly optimizes cluster assignments with a soft statistical-parity regularizer. A single parameter $\lambda$ tunes the fairness--utility balance, while nonnegativity yields parts-based factors and transparent soft memberships. The optimization uses sparse-friendly alternating updates and scales near-linearly with the number of edges. Across synthetic and real networks, DFNMF achieves substantially higher group balance at comparable modularity, often dominating state-of-the-art baselines on the Pareto front. The code is available at https://github.com/SiamakGhodsi/DFNMF.git.

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 / 1 minor

Summary. The manuscript proposes DFNMF, an end-to-end deep nonnegative tri-factorization model for fair graph clustering. It augments standard NMF-style factorization with a soft statistical-parity regularizer whose strength is controlled by a single scalar λ, and optimizes the joint objective via sparse-friendly alternating updates. The central empirical claim is that DFNMF produces substantially higher group balance at comparable modularity than existing baselines and frequently dominates the Pareto front on both synthetic and real networks.

Significance. If the reported Pareto dominance is robust, the work would supply a scalable, parts-based, and single-parameter method for trading off community structure against demographic parity in graph clustering—an improvement over rigid-constraint or multi-stage pipelines. The public code release is a positive factor for verification.

major comments (3)
  1. [Abstract and Experiments] The abstract and experimental section report superior Pareto performance on synthetic and real networks, yet supply no quantitative details on the exact baselines, number of runs, error bars, statistical significance tests, or data-exclusion rules; without these the dominance claim cannot be evaluated.
  2. [Optimization and Experiments] The joint objective is non-convex; the manuscript does not report results across multiple random initializations or iteration-order permutations, leaving open the possibility that the observed fairness-utility trade-off is an artifact of favorable convergence rather than a stable property of the model class (see skeptic note on alternating optimization).
  3. [Method and Experiments] The claim that a single λ produces a controllable and generalizable trade-off rests on the assumption that the soft parity regularizer interacts cleanly with the tri-factorization; no ablation isolating the regularizer’s effect or testing sensitivity to initialization is provided.
minor comments (1)
  1. [Method] Notation for the tri-factorization matrices and the precise form of the statistical-parity term should be stated explicitly in the main text rather than deferred to the appendix.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thoughtful and constructive comments. These points help clarify the experimental rigor and robustness of our claims. We address each major comment below and commit to revisions that strengthen the manuscript without altering its core contributions.

read point-by-point responses

  1. Referee: [Abstract and Experiments] The abstract and experimental section report superior Pareto performance on synthetic and real networks, yet supply no quantitative details on the exact baselines, number of runs, error bars, statistical significance tests, or data-exclusion rules; without these the dominance claim cannot be evaluated.

    Authors: We agree that the current presentation lacks sufficient experimental detail to allow full evaluation of the Pareto-dominance claims. In the revised manuscript we will add: (i) an explicit table listing all baselines with citations, (ii) the precise number of independent runs (we will report results over 10 random seeds), (iii) error bars as mean ± one standard deviation, (iv) statistical significance results using paired Wilcoxon signed-rank tests against the strongest baseline, and (v) a clear description of data preprocessing and any exclusion criteria applied to the real-world networks. These additions will be placed in a dedicated “Experimental Setup” subsection. revision: yes

  2. Referee: [Optimization and Experiments] The joint objective is non-convex; the manuscript does not report results across multiple random initializations or iteration-order permutations, leaving open the possibility that the observed fairness-utility trade-off is an artifact of favorable convergence rather than a stable property of the model class (see skeptic note on alternating optimization).

    Authors: We acknowledge the non-convex nature of the joint objective and the potential sensitivity of alternating optimization to initialization and update ordering. Although our internal checks indicated stable convergence behavior, we did not systematically document this in the original submission. In the revision we will add a new subsection reporting performance across 10 distinct random initializations for both synthetic and real networks, together with a brief analysis of iteration-order sensitivity. We will also include a short discussion of why the observed trade-off appears robust under the chosen sparse-friendly alternating scheme. revision: yes

  3. Referee: [Method and Experiments] The claim that a single λ produces a controllable and generalizable trade-off rests on the assumption that the soft parity regularizer interacts cleanly with the tri-factorization; no ablation isolating the regularizer’s effect or testing sensitivity to initialization is provided.

    Authors: We appreciate this observation. To substantiate the controllability claim, the revised manuscript will contain an ablation study that directly compares DFNMF with the regularizer disabled (λ = 0) against the full model across a range of λ values. We will also report sensitivity of the fairness–utility frontier to different random initializations and provide a brief theoretical note on how the soft parity term interacts with the nonnegative tri-factorization constraints. These additions will be supported by additional figures and tables. revision: yes

Circularity Check

0 steps flagged

No significant circularity; method introduces independent regularizer and optimization

full rationale

The paper presents DFNMF as a new end-to-end deep nonnegative tri-factorization model that adds a soft statistical-parity regularizer (controlled by a single tunable lambda) to standard NMF-style factorization for graphs. The alternating sparse-friendly updates and Pareto-front claims are derived from this joint objective rather than reducing to any pre-fitted quantity, self-citation chain, or definitional equivalence within the paper's own equations. No load-bearing step quotes or relies on prior author work as an unverified uniqueness theorem or ansatz; results are framed as empirical outcomes on synthetic and real networks. The derivation remains self-contained against external NMF baselines plus the novel fairness term.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard nonnegative matrix factorization assumptions plus the new regularizer; the main free parameter is the user-chosen trade-off weight.

free parameters (1)
  • lambda
    Single scalar that controls the strength of the fairness regularizer relative to the clustering objective; chosen by the user rather than learned from data.
axioms (2)
  • domain assumption Nonnegativity of the latent factors produces parts-based representations and transparent soft cluster memberships.
    Invoked to justify interpretability of the tri-factorization output.
  • domain assumption Alternating sparse-friendly updates converge to a useful solution for the joint clustering-plus-fairness objective on real graphs.
    Underlies the claimed near-linear scaling and practical performance.

pith-pipeline@v0.9.0 · 5728 in / 1380 out tokens · 53116 ms · 2026-05-21T20:21:23.864601+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages

  1. [1]

    Bias in data-driven artificial intelligence systems: An introductory survey,

    E. Ntoutsi, P. Fafalios, U. Gadiraju, V . Iosifidis, W. Nejdl, M.-E. Vidal, S. Ruggieri, F. Turini, S. Papadopoulos, E. Krasanakiset al., “Bias in data-driven artificial intelligence systems: An introductory survey,” Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, vol. 10, no. 3, p. e1356, 2020

    work page 2020

  2. [2]

    A survey on datasets for fairness-aware machine learning,

    T. L. Quy, A. Roy, V . Iosifidis, W. Zhang, and E. Ntoutsi, “A survey on datasets for fairness-aware machine learning,”WIREs Data Mining Knowl. Discov., vol. 12, no. 3, 2022

    work page 2022

  3. [3]

    Adversarial reweighting guided by wasserstein distance to achieve demographic parity,

    X. Zhao, S. Fabbrizzi, P. R. Lobo, S. Ghodsi, K. Broelemann, S. Staab, and G. Kasneci, “Adversarial reweighting guided by wasserstein distance to achieve demographic parity,” inIEEE Big Data. IEEE, 2024, pp. 1605–1614

    work page 2024

  4. [4]

    Verifying cross-modal entity consistency in news using vision-language models,

    S. Tahmasebi, E. Müller-Budack, and R. Ewerth, “Verifying cross-modal entity consistency in news using vision-language models,” inECIR (4), ser. Lecture Notes in Computer Science, vol. 15575. Springer, 2025, pp. 339–354

    work page 2025

  5. [5]

    Fairness in graph mining: A survey,

    Y . Dong, J. Ma, S. Wang, C. Chen, and J. Li, “Fairness in graph mining: A survey,”IEEE Trans. Knowl. Data Eng., vol. 35, no. 10, pp. 10 583– 10 602, 2023

    work page 2023

  6. [6]

    Policy advice and best practices on bias and fairness in AI,

    J. M. Álvarez, A. B. Colmenarejo, A. Elobaid, S. Fabbrizzi, M. Fahimi, A. Ferrara, S. Ghodsi, C. Mougan, I. Papageorgiou, P. R. Lobo, M. Russo, K. M. Scott, L. State, X. Zhao, and S. Ruggieri, “Policy advice and best practices on bias and fairness in AI,”Ethics Inf. Technol., vol. 26, no. 2, p. 31, 2024

    work page 2024

  7. [7]

    A survey of data-efficient graph learning,

    W. Ju, S. Yi, Y . Wang, Q. Long, J. Luo, Z. Xiao, and M. Zhang, “A survey of data-efficient graph learning,” inIJCAI, 2024, pp. 8104–8113

    work page 2024

  8. [8]

    Multi-fair capacitated students- topics grouping problem,

    T. L. Quy, G. Friege, and E. Ntoutsi, “Multi-fair capacitated students- topics grouping problem,” inPAKDD (1), ser. LNCS, vol. 13935. Springer, 2023, pp. 507–519

    work page 2023

  9. [9]

    Fair clustering through fairlets,

    F. Chierichetti, R. Kumar, S. Lattanzi, and S. Vassilvitskii, “Fair clustering through fairlets,” inNeurIPS, 2017, pp. 5029–5037

    work page 2017

  10. [10]

    A scalable algorithm for individually fair k-means clustering,

    M. Bateni, V . Cohen-Addad, A. Epasto, and S. Lattanzi, “A scalable algorithm for individually fair k-means clustering,” inAISTATS, vol. 238. PMLR, 2024, pp. 3151–3159

    work page 2024

  11. [11]

    Individual fairness for graph neural networks: A ranking based approach,

    Y . Dong, J. Kang, H. Tong, and J. Li, “Individual fairness for graph neural networks: A ranking based approach,” inKDD. ACM, 2021, pp. 300–310

    work page 2021

  12. [12]

    Fairness-aware graph neural networks: A survey,

    A. Chen, R. A. Rossi, N. Park, P. Trivedi, Y . Wang, T. Yu, S. Kim, F. Dernoncourt, and N. K. Ahmed, “Fairness-aware graph neural networks: A survey,”ACM Trans. Knowl. Discov. Data, vol. 18, no. 6, pp. 138:1– 138:23, 2024

    work page 2024

  13. [13]

    Fairsin: Achieving fairness in graph neural networks through sensitive information neutralization,

    C. Yang, J. Liu, Y . Yan, and C. Shi, “Fairsin: Achieving fairness in graph neural networks through sensitive information neutralization,” inAAAI. AAAI Press, 2024, pp. 9241–9249

    work page 2024

  14. [14]

    Guarantees for spectral clustering with fairness constraints,

    M. Kleindessner, S. Samadi, P. Awasthi, and J. Morgenstern, “Guarantees for spectral clustering with fairness constraints,” inICML, vol. 97, 2019, pp. 3458–3467

    work page 2019

  15. [15]

    Consistency of constrained spectral clustering under graph induced fair planted partitions,

    S. Gupta and A. Dukkipati, “Consistency of constrained spectral clustering under graph induced fair planted partitions,” inNeurIPS, 2022, pp. 13 527–13 540

    work page 2022

  16. [16]

    Scalable spectral clustering with group fairness constraints,

    J. Wang, D. Lu, I. Davidson, and Z. Bai, “Scalable spectral clustering with group fairness constraints,” inAISTATS, 2023, pp. 6613–6629

    work page 2023

  17. [17]

    Towards cohesion-fairness harmony: Contrastive regularization in individual fair graph clustering,

    S. Ghodsi, S. A. Seyedi, and E. Ntoutsi, “Towards cohesion-fairness harmony: Contrastive regularization in individual fair graph clustering,” inPAKDD (1), vol. 14645, 2024, pp. 284–296

    work page 2024

  18. [18]

    Graph learning based recommender systems: A review,

    S. Wang, L. Hu, Y . Wang, X. He, Q. Z. Sheng, M. A. Orgun, L. Cao, F. Ricci, and P. S. Yu, “Graph learning based recommender systems: A review,” inProceedings of the 30th IJCAI, 2021, pp. 4644–4652

    work page 2021

  19. [19]

    Affinity clustering framework for data debiasing using pairwise distribution discrepancy,

    S. Ghodsi and E. Ntoutsi, “Affinity clustering framework for data debiasing using pairwise distribution discrepancy,” inEWAF, ser. CEUR Proceedings, vol. 3442, 2023

    work page 2023

  20. [20]

    Context matters for fairness - a case study on the effect of spatial distribution shifts,

    S. Ghodsi, H. Alani, and E. Ntoutsi, “Context matters for fairness - a case study on the effect of spatial distribution shifts,”CoRR, vol. abs/2206.11436, 2022

    work page arXiv 2022

  21. [21]

    SoK: Fair Clustering: Critique, Caveats, and Future Directions,

    J. Dickerson, S. A. Esmaeili, J. Morgenstern, and C. J. Zhang, “SoK: Fair Clustering: Critique, Caveats, and Future Directions,” inIEEE Conference on Secure and Trustworthy Machine Learning (SaTML). IEEE Computer Society, Apr. 2025, pp. 698–713

    work page 2025

  22. [22]

    Spectral normalized-cut graph partitioning with fairness constraints,

    J. Li, Y . Wang, and A. Merchant, “Spectral normalized-cut graph partitioning with fairness constraints,” inECAI, ser. Frontiers in Artificial Intelligence and Applications, vol. 372. IOS Press, 2023, pp. 1389–1397

    work page 2023

  23. [23]

    Graph clustering with graph neural networks,

    A. Tsitsulin, J. Palowitch, B. Perozzi, and E. Müller, “Graph clustering with graph neural networks,”J. Mach. Learn. Res., vol. 24, pp. 127:1– 127:21, 2023

    work page 2023

  24. [24]

    Learning the parts of objects by non- negative matrix factorization,

    D. D. Lee and H. S. Seung, “Learning the parts of objects by non- negative matrix factorization,”Nature, vol. 401, no. 6755, pp. 788–791, 1999

    work page 1999

  25. [25]

    Gillis,Nonnegative matrix factorization

    N. Gillis,Nonnegative matrix factorization. SIAM, 2020

    work page 2020

  26. [26]

    Self- paced multi-label learning with diversity,

    S. A. Seyedi, S. S. Ghodsi, F. A. Tab, M. Jalili, and P. Moradi, “Self- paced multi-label learning with diversity,” inACML, ser. Proceedings of Machine Learning Research, vol. 101. PMLR, 2019, pp. 790–805

    work page 2019

  27. [27]

    Symmetric nonnegative matrix factorization for graph clustering,

    D. Kuang, H. Park, and C. H. Q. Ding, “Symmetric nonnegative matrix factorization for graph clustering,” inSDM, 2012, pp. 106–117

    work page 2012

  28. [28]

    Nonnegative matrix tri- factorization with graph regularization for community detection in social networks,

    Y . Pei, N. Chakraborty, and K. P. Sycara, “Nonnegative matrix tri- factorization with graph regularization for community detection in social networks,” inIJCAI. AAAI Press, 2015, pp. 2083–2089

    work page 2015

  29. [29]

    Deep asymmetric nonnegative matrix factorization for graph clustering,

    A. Hajiveiseh, S. A. Seyedi, and F. A. Tab, “Deep asymmetric nonnegative matrix factorization for graph clustering,”Pattern Recognit., vol. 148, p. 110179, 2024

    work page 2024

  30. [30]

    A deep semi-nmf model for learning hidden representations,

    G. Trigeorgis, K. Bousmalis, S. Zafeiriou, and B. W. Schuller, “A deep semi-nmf model for learning hidden representations,” inICML, vol. 32. JMLR.org, 2014, pp. 1692–1700

    work page 2014

  31. [31]

    A survey on deep matrix factorizations,

    P. D. Handschutter, N. Gillis, and X. Siebert, “A survey on deep matrix factorizations,”Comput. Sci. Rev., vol. 42, p. 100423, 2021

    work page 2021

  32. [32]

    A tutorial on spectral clustering,

    U. von Luxburg, “A tutorial on spectral clustering,”Statistics and Computing, vol. 17, no. 4, pp. 395–416, 2007

    work page 2007

  33. [33]

    Metrics for community analysis: A survey,

    T. Chakraborty, A. Dalmia, A. Mukherjee, and N. Ganguly, “Metrics for community analysis: A survey,”ACM Computing Surveys, vol. 50, no. 4, pp. 1–37, 2017

    work page 2017

  34. [34]

    Fairness definitions explained,

    S. Verma and J. Rubin, “Fairness definitions explained,” inFair- Ware@ICSE. ACM, 2018, pp. 1–7

    work page 2018

  35. [35]

    Say no to the discrimination: Learning fair graph neural networks with limited sensitive attribute information,

    E. Dai and S. Wang, “Say no to the discrimination: Learning fair graph neural networks with limited sensitive attribute information,” inWSDM, 2021, pp. 680–688

    work page 2021

  36. [36]

    Characteristic functions on graphs: Birds of a feather, from statistical descriptors to parametric models,

    B. Rozemberczki and R. Sarkar, “Characteristic functions on graphs: Birds of a feather, from statistical descriptors to parametric models,” in CIKM, 2020, pp. 1325–1334

    work page 2020

  37. [37]

    Social networks of drug users in high-risk sites: Finding the connections,

    M. R. Weeks, S. Clair, S. P. Borgatti, K. Radda, and J. J. Schensul, “Social networks of drug users in high-risk sites: Finding the connections,”AIDS and Behavior, vol. 6, pp. 193–206, 2002

    work page 2002

  38. [38]

    Contact patterns in a high school: a comparison between data collected using wearable sensors, contact diaries and friendship surveys,

    R. Mastrandrea, J. Fournet, and A. Barrat, “Contact patterns in a high school: a comparison between data collected using wearable sensors, contact diaries and friendship surveys,”PloS one, vol. 10, no. 9, p. e0136497, 2015

    work page 2015

  39. [39]

    Community discovery using nonnegative matrix factorization,

    F. Wang, T. Li, X. Wang, S. Zhu, and C. H. Q. Ding, “Community discovery using nonnegative matrix factorization,”Data Min. Knowl. Discov., vol. 22, no. 3, pp. 493–521, 2011

    work page 2011

  40. [40]

    An ideal point based many-objective optimization for community detection of complex networks,

    S. Tahmasebi, P. Moradi, S. Ghodsi, and A. Abdollahpouri, “An ideal point based many-objective optimization for community detection of complex networks,”Inf. Sci., vol. 502, pp. 125–145, 2019

    work page 2019

  41. [41]

    Many-objective evolutionary optimization using density peaks scoring selection strategy,

    S. Ghodsi, S. Tahmasebi, M. Jalili, and P. Moradi, “Many-objective evolutionary optimization using density peaks scoring selection strategy,” inGECCO Companion. ACM, 2024, pp. 331–334

    work page 2024

  42. [42]

    On the convergence of multiplicative update algorithms for nonnegative matrix factorization,

    C. Lin, “On the convergence of multiplicative update algorithms for nonnegative matrix factorization,”IEEE Trans. Neural Networks, vol. 18, no. 6, pp. 1589–1596, 2007

    work page 2007

  43. [43]

    Rethinking neural vs. matrix-factorization collaborative filtering: the theoretical perspectives,

    D. Xu, C. Ruan, E. Körpeoglu, S. Kumar, and K. Achan, “Rethinking neural vs. matrix-factorization collaborative filtering: the theoretical perspectives,” inICML, vol. 139, 2021, pp. 11 514–11 524

    work page 2021

  44. [44]

    Neural collaborative filtering vs. matrix factorization revisited,

    S. Rendle, W. Krichene, L. Zhang, and J. R. Anderson, “Neural collaborative filtering vs. matrix factorization revisited,” inRecSys. ACM, 2020, pp. 240–248

    work page 2020

  45. [45]

    Fairbranch: Mitigating bias transfer in fair multi-task learning,

    A. Roy, C. Koutlis, S. Papadopoulos, and E. Ntoutsi, “Fairbranch: Mitigating bias transfer in fair multi-task learning,” inIJCNN. IEEE, 2024, pp. 1–8

    work page 2024

  46. [46]

    Multi-dimensional discrimination in law and machine learning - A comparative overview,

    A. Roy, J. Horstmann, and E. Ntoutsi, “Multi-dimensional discrimination in law and machine learning - A comparative overview,” inFAccT. ACM, 2023, pp. 89–100

    work page 2023

  47. [47]

    Mmm-fair: An interactive toolkit for exploring and operationalizing multi-fairness trade- offs,

    S. Swati, A. Roy, E. Panagiotou, and E. Ntoutsi, “Mmm-fair: An interactive toolkit for exploring and operationalizing multi-fairness trade- offs,”CoRR, vol. abs/2509.08156, 2025. VIII. APPENDIX1: THEORETICS ANDDIFFERENTIAL CALCULUS A. Epistemic Comparison of NMF & DNN In this section, we discuss theoretical and epistemic differ- ences between NMF-based an...

    work page arXiv 2025