Pith. sign in

REVIEW 4 major objections 5 minor 38 references

Fast In-Spectrum Graph Watermarks

T0 review · 4 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read The paper claims that graph watermarking can be made practical for very large graphs by embedding a Gaussian key into the Fourier spectrum of the adjacency matrix and binarizing the result, yielding $O(N^2 \log N)$ time and robustness to…

desk verdict Porting Cox spectral image watermarking to adjacency matrices is a genuine new idea with real complexity gains, but the robustness claims hinge on an untested label-fixed assumption and fitted thresholds. read the letter →

arxiv 2502.04182 v2 pith:CYV25TQU submitted 2025-02-06 cs.DS

Jade Garcia Bourrée , Anne-Marie Kermarrec , Erwan Le Merrer , Othmane Safsafi This is my paper · ORCID
classification cs.DS
keywords graphwatermarkingFouriertransformadjacencymatrixspreadspectrumprovenancelargegraphsedgeflipattackbinarization
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

This paper claims that graph watermarking can be made practical for very large graphs by treating the adjacency matrix as an image and embedding the watermark in its Fourier spectrum. The resulting scheme, F&F, runs in $O(N^2 \log N)$ time for an $N$-vertex graph, far below the NP-complete complexity of the two existing graph-watermarking methods. If the claim holds, owners of million-vertex graphs can embed and verify provenance efficiently, with robustness to random edge flips that matches or exceeds the state of the art.

What carries the argument

The scheme's central object is the discrete Fourier transform of the binary adjacency matrix $A$. A Gaussian key of length $m$ is added to the $m$ smallest-amplitude Fourier coefficients, the inverse transform is applied, and the result is binarized by thresholding at the average value of $A$, producing the watermarked adjacency matrix $A_W$. Because this binarization spreads the key across all coefficients, extraction compares the full spectral difference between the original and suspect matrices rather than only the $m$ modified locations.

What would settle it

Take a watermarked graph, apply a random permutation to its vertex labels, and run the extraction: if the watermark is unrecoverable after this relabeling while an edge-flip attack of the same magnitude is survived, the scheme's robustness claim fails for relabeling attacks. A second decisive test is to measure the false-positive rate on graphs with density below 2, where the paper itself reports collisions between distinct keys.

Watch

Extended reading notes

Core claim

The central claim is that the in-spectrum watermarking technique of Cox et al., originally designed for images, transfers to unweighted undirected graphs once the adjacency matrix is treated as a pixel matrix and the watermarked spectrum is binarized back to $\{0,1\}$. The authors prove that an unattacked watermarked graph always passes extraction when the threshold is zero, derive a closed-form probability that two independent keys produce distinct watermarked graphs, and report experimental resilience to edge flips beyond 100% of the edge count on several million-vertex real graphs. In a head-to-head benchmark, F&F matches or beats the subgraph-matching scheme of Zhao et al. and the isomorphism-based scheme of Eppstein et al. in robustness while being fast enough to watermark graphs an order of magnitude or more larger within the same timeout.

Load-bearing premise

The scheme requires that the suspect graph's vertices carry the same fixed labels used at embedding time, so that its adjacency matrix lines up entry-by-entry with the watermarked one; if a graph is relabeled or its vertex order is unknown, the Fourier comparison cannot be aligned and the watermark cannot be found.

Editorial extensions

If this is right

  • Graph owners can embed provenance watermarks into graphs with millions of vertices in minutes rather than running intractable subgraph searches.
  • The scheme withstands random edge-flip attacks at least up to 100% of the edge count, an intensity the paper argues already destroys the attacked graph's utility.
  • Extraction time does not grow with the intensity of the attack, so F&F's extraction is effectively constant for large graphs, unlike matching-based methods.
  • The threshold $\theta$ can be chosen per graph model and density so that no false positives occur for $\theta \leq 5$ on Barabási-Albert graphs.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the scheme is not invariant to vertex relabeling, an adversary who permutes labels could strip the watermark; testing this attack would likely expose the main practical gap.
  • The binarization step limits capacity: key length and $\sigma$ must be retuned for each graph density, so transferring the scheme to directed, weighted, or extremely sparse graphs is not automatic.
  • The reported survival of 100% edge flips is surprising and likely reflects the watermark's spread across the whole spectrum; a theoretical account of this robustness would strengthen the claim.
  • The undetectability result was obtained only on Barabási-Albert graphs, so it remains open whether the watermark is equally invisible in other graph families.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes F&F, a graph watermarking scheme that embeds a Gaussian key into selected Fourier coefficients of a graph's adjacency matrix, then binarizes the inverse transform to recover an unweighted graph. Extraction compares the Fourier-domain difference between the original and suspected graph against a threshold. The authors claim an O(N^2 log N) complexity (or O(N0^2 log N0) after dimensionality reduction), and present experiments on synthetic and large real-world graphs to support low distortion, uniqueness, low false positives/negatives, robustness to edge flips, and runtime advantages over the subgraph-based schemes of Zhao et al. and Eppstein et al.

Significance. If the central claims held, the paper would make an important practical contribution: it would reduce graph watermarking from NP-hard subgraph/isomorphism-based methods to fast Fourier operations, and it would scale to graphs with millions of vertices, as demonstrated on several SNAP and Network Repository instances. The paper deserves credit for a simple, reproducible core construction, for benchmarking against the two main baselines on real graphs, and for extending the evaluation beyond edge flips to a GNN-based undetectability test. However, the significance is conditional on the coordinate-locked labeling assumption and on threshold parameters that are fitted to produce the reported behavior; both issues affect the head-to-head comparison and the robustness claims.

major comments (4)
  1. [Section III, Algorithm 1] Algorithm 1's Embed function binarizes A' but never symmetrizes the result, even though the paper states in Section III that binarization and symmetrization are required to return an undirected graph and Figure 7 depicts a symmetric adjacency matrix. As written, line 11 can produce a directed binary matrix, making the watermarked graph ill-defined and the edit distances reported in Tables I and II inconsistent.
  2. [Appendix C, Section IV-C] The binarization threshold uses av(A) = |E|/N^2, but for an undirected graph with zero diagonal the average of the adjacency matrix is 2|E|/N^2. This factor of two propagates into Equation (2), the density-collision discussion, and the claim that densities below 2 are atypical; the quantitative uniqueness conclusions are therefore not supported by the stated formula.
  3. [Section III and Section IV-B2] Extraction is coordinate-locked: Section III assumes a deterministic vertex labeling, and the proof at the end of Section IV-B2 uses A* = A_W, which holds only when the suspected graph is presented in the same labeling. For a relabeled watermarked graph G* = P A_W P^T, the matrix A - A* is not a sparse edge difference and the norm criterion s <= theta ||W||_2 has no invariance. Since both baselines are subgraph/isomorphism-based and hence label-invariant, the abstract's claim of performing 'at least as well' is tested only in a coordinate system that favors F&F; no experiment applies a vertex permutation.
  4. [Section IV-C and Section IV-F3] The key parameters sigma and theta are chosen by dichotomous search to achieve the target distortion and the target attack resistance (Section IV-C; Figure 5 and Table II). Consequently, the reported robustness is a property of the fitted thresholds rather than an independent prediction of the scheme. The statement in Section IV-B2 that setting theta to tolerate 10% edge flips guarantees no false negatives below 10% flips is an assertion by construction, not a demonstrated property over graph distributions, and the false-positive study in Figure 3 covers only one generator (BA) at one density.
minor comments (5)
  1. [Section IV-A, Definition of edit distance] The definition says ED is the percentage of edges that distinguish two graphs, but the examples and figures use ED as a ratio (e.g., 100% flips, 1000% of flips in Section V-C); please make the percentage-versus-ratio convention consistent throughout.
  2. [Section V-B, Figure 6a] The text states that F&F extraction is 'almost constant after 10k vertices,' but the log-scale plot and the O(N^2 log N) complexity do not support a constant asymptotic claim; please clarify whether this refers to measured wall-clock time on a specific implementation rather than complexity.
  3. [Appendix D, Table IV] The adapted Embed complexity is given as O((N + N0^2) log N0), but selecting the top N0 highest-degree vertices from N vertices requires at least O(N log N) or O(N + N0 log N) operations if done by partial sorting; please specify the selection algorithm and its cost.
  4. [Table II] The row for kron-g500-logn20 lists m = 71128, which appears inconsistent with the other key lengths and with the claim that m follows Eppstein et al.'s density-based guideline; please double-check this value and the associated threshold.
  5. [Figure 5] The x-axis of Figure 5 is labeled 'ED' in the caption, but the text describes varying the percentage of edge flips for which theta is calibrated; please correct the labeling to avoid confusing edit distance with flip percentage.

Circularity Check

3 steps flagged · score 6.0 of 10

Robustness and low-distortion results are partly self-fulfilling: θ and σ are fitted to the exact attack and edit-distance targets that are then reported as achievements; the O(N^2 log N) complexity analysis is independent.

  1. fitted input called prediction [Section IV-B2, paragraph "A realistic case with an attacked watermarked graph"]
    "We set θ to tolerate 10% edge flips, which is a conservatively high value (according to related work [13], [38], as it will be illustrated in Section IV-C). Consequently, no false negatives can occur with less than 10 % edge flips."

    The extraction success criterion is s = ||W* − W||_2 ≤ θ ||W||_2. θ is chosen specifically so that this inequality holds for graphs with 10% of edges flipped. The sentence "Consequently, no false negatives can occur with less than 10% edge flips" is therefore a restatement of the fitting target, not an independent robustness measurement. The threshold is the free parameter being tuned to the exact attack strength, so the headline no-false-negative claim reduces to the chosen value of θ.

  2. fitted input called prediction [Section IV-F3, "Densities and the Resulting Threshold"]
    "For each generator and each density, we set θ to be resistant to x % edge flips ( x-axis). This resilience is observed in 3 runs per point on Figure 5, where a dichotomous search sets θ to achieve the effective resilience to the attack."

    The paper states explicitly that θ is set to be resistant to the attack and that a dichotomous search "sets θ to achieve the effective resilience to the attack." The plotted resilience is thus the optimization objective of the search over θ, not a predictive property of the scheme evaluated at a fixed, independently chosen parameter. Reporting this curve as evidence for Goal 4 presents the fitted parameter as the experimental result.

1 more flagged steps
  1. fitted input called prediction [Section IV-C, "(Goal 1) Low Distortion"]
    "An automated way to set σ under the small ED constraint is to use a dichotomous search in [ [1, σmax] ]... Applying this method to the three significant real graphs results in the parameters listed in Table I. The observed edit distances remain well under the 10−2% target, fulfilling Goal 1."

    The low-distortion outcome is obtained by searching σ until the edit distance is below the target. The "observed edit distances" are the stopping condition of the dichotomous search, so the claim that F&F fulfills Goal 1 is a property of the tuning procedure rather than an independent consequence of the watermarking scheme. This is the same fitted-input pattern as the θ selection, though it is a design-goal demonstration rather than a prediction.

full rationale

The paper's core algorithmic contribution—FFT-based embedding and extraction with O(N^2 log N) worst-case complexity—is derived from standard Fourier-transform and sorting bounds and is not circular. The no-attack θ=0 proof is a valid but definitional identity: when A* = A_W, linearity gives W* = W, so the success inequality becomes 0 ≤ 0. The circularity burden lies in the robustness evaluation. In Section IV-B2, θ is set specifically to tolerate 10% edge flips and the paper then states that no false negatives can occur under 10% flips; this is a restatement of the fitting target. In Section IV-F3, a dichotomous search "sets θ to achieve the effective resilience to the attack," so the plotted resilience is the optimization objective, not an independent measurement. The low-distortion procedure similarly searches σ until the edit distance falls under the 10^-2% target and then reports that target as fulfilled. These are partial circularities: they show the scheme can be tuned to meet its own success criteria, but they do not independently predict robustness or distortion. The observation that extraction succeeds even with 100% flipped edges, using a θ fitted to 10%, is genuine evidence beyond the fitted target, but it does not rescue the fitted "no false negatives below 10%" statements. The comparison with baselines also inherits the coordinate-locked extraction assumption, since Extract compares A − A* entrywise in the owner's fixed labeling; that is a correctness and scope limitation rather than a circular step. Overall, the complexity claim and the basic scheme are self-contained, but the headline robustness and low-distortion claims are partly driven by fitted thresholds, giving a score of 6.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central scheme rests on standard Fourier linearity plus several domain assumptions. The most fragile are the deterministic-labeling assumption, which makes the method sensitive to vertex permutations, and the empirical assertions that sigma_max exists and that theta can be calibrated to guarantee false-negative freedom. The uniqueness guarantee also excludes low-density graphs based on observed collisions rather than a proven threshold. No new physical entities are introduced.

free parameters (4)
  • m (key length) = e.g., 210 (BA), 3,250 (Flickr), 170 (Pokec); 54 to 71,128 in Table II
    Chosen per graph density following Eppstein et al. [13]; affects distortion and uniqueness; not derived from first principles.
  • sigma (Gaussian key standard deviation) = e.g., 1,750 (BA), 32,000 (Flickr), 7,000 (Pokec); 61 to 224,000 in Table II
    Tuned by dichotomous search to keep edit distance under 10^-2 percent; a fitted knob that controls watermark energy.
  • theta (extraction threshold) = 0.6 to 374 in Table II; theta=1.4 for BA density 5; theta=1 for BA/WS at 10% flips
    Calibrated via experiments and dichotomous search to tolerate up to 10% edge flips; robustness results depend on this fitted value.
  • N0 (dimensionality reduction size) = 10,000
    Hand-chosen number of highest-degree vertices; watermark is only embedded in this subgraph, so attacks outside it are ignored.
assumptions (6)
  • standard math FFT is linear and invertible; linearity lets the watermark be added in the spectral domain equivalently to the spatial domain.
    Used in Algorithm 1 lines 7-10 and in the proof for theta=0 in Section IV-B2.
  • domain assumption A deterministic vertex ordering exists, so adjacency matrices of the original and suspected graph are aligned coordinate-wise.
    Stated in Section III: 'we can label the input graph arbitrarily'; without it, Extract cannot compare matrices.
  • domain assumption The top N0 highest-degree vertices contain the significant graph information, analogous to low frequencies in images.
    Scalability reduction in Appendix D; if an attacker modifies low-degree vertices outside this subgraph, the watermark is unaffected but the subgraph assumption defines what is protected.
  • ad hoc to paper Graphs with density below 2 are rare or atypical, and their watermark collisions can be excluded from the uniqueness guarantee.
    Section IV-D dismisses observed collisions for low densities instead of providing a theoretical bound; this restricts Goal 2 without proof.
  • ad hoc to paper For any graph and key length m, a sigma_max exists that yields a nonzero edit distance, and sigma can be found by dichotomous search.
    Appendix D asserts this from 'experiments have shown', not from a proof.
  • ad hoc to paper Threshold theta calibrated to tolerate 10% edge flips guarantees no false negatives for attacks below 10% flips.
    Section IV-B2 states this without a derivation linking edit distance to the extraction norm; the claim is empirical calibration.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fast In-Spectrum Graph Watermarks." pith.science (2026). https://pith.science/paper/CYV25TQU

@misc{pith2026250204182,
  author       = {Pith},
  title        = {Pith review of: Fast In-Spectrum Graph Watermarks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CYV25TQU}},
  note         = {Machine review of arXiv:2502.04182}
}
abstract

We address the problem of watermarking graph objects, which consists in hiding information within them, to prove their origin. The two existing methods to watermark graphs use subgraph matching or graph isomorphism techniques, which are known to be intractable for large graphs. To reduce the operational complexity, we propose FFG, a new graph watermarking scheme adapted from an image watermarking scheme, since graphs and images can be represented as matrices. We analyze and compare FFG, whose novelty lies in embedding the watermark in the Fourier transform of the adjacency matrix of a graph. Our technique enjoys a much lower complexity than that of related works (i.e. in $\mathcal{O}\left(N^2 \log N\right)$), while performing better or at least as well as the two state-of-the-art methods.

Figures

Figures reproduced from arXiv: 2502.04182 by the authors.

Figure 1
Figure 1. illustrates the link between σ and m for our three generative graph models, and a slight distortion of 0.005% edges. Key lengths used in [13] depend on densities (being defined as the number of edges over the number of vertices) [26], while those of [38] depend on the number of vertices. We set the key length m of F&F as the key length used in [13] because the experiment in Section IV-D shows that F&F is also relate… view at source ↗
Figure 2
Figure 2. Success rate (y-axis) in inserting a unique watermark in three graph models, for 1M vertices graph models and different densities on the x-axis. E. (Goal 3) Low False Positives and Negatives We now study the effect of θ on false positives. Let GW be the watermarked graph of G with the key ω, and G∗ be any other graph not watermarked with ω. Recall that Algorithm 1 with G∗ results in a false positive if the extractio… view at source ↗
Figure 3
Figure 3. The impact of θ on extraction false positives (BA graphs). experiment in [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: The effect of an edge flipping attack on three real [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: The choice of θ in F&F to be resistant to x% edge flips depending on the density and the type of graph model. V. A COMPARATIVE BENCHMARK Finally, we perform a head-to-head comparative benchmark of F&F with the two state-of-the-art approaches [38] (Zhao) and [13] (Eppst…
Figure 6
Figure 6. Figure 6: Benchmark comparison between Scheme, Epp [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: The F&F framework for watermarking unweighted [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 35 canonical work pages

  1. [1]

    Watermarking relational databases

    Rakesh Agrawal and Jerry Kiernan. Watermarking relational databases. In VLDB’02: Proceedings of the 28th International Conference on Very Large Databases, pages 155–166. Elsevier, 2002

    work page 2002

  2. [2]

    Graph spectral domain blind watermarking

    Hiba Al-Khafaji and Charith Abhayaratne. Graph spectral domain blind watermarking. In ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages 2492– 2496, 2019

    work page 2019

  3. [3]

    Special issue on big graph data management and processing

    Angela Bonifati and Hannes V oigt. Special issue on big graph data management and processing. The VLDB Journal, 31(2):201–202, 2022

    work page 2022

  4. [4]

    Secure spread spectrum watermarking for multimedia

    Ingemar Johansson Cox, Joe Kilian, Frank Thomson Leighton, and Talal Shamoon. Secure spread spectrum watermarking for multimedia. IEEE transactions on image processing , 6(12):1673–1687, 1997

    work page 1997

  5. [5]

    Graph data management systems for new application domains

    Philippe Cudr ´e-Mauroux and Sameh Elnikety. Graph data management systems for new application domains. Proceedings of the VLDB Endowment, 4(12):1510–1511, 2011

    work page 2011

  6. [6]

    Study on centrality measures in social networks: a survey

    Kousik Das, Sovan Samanta, and Madhumangal Pal. Study on centrality measures in social networks: a survey. Social network analysis and mining, 8(1):1–11, 2018

    work page 2018

  7. [7]

    A tale of two graphs: Property graphs as rdf in oracle

    Souripriya Das, Jagannathan Srinivasan, Matthew Perry, Eugene Inseok Chong, and Jayanta Banerjee. A tale of two graphs: Property graphs as rdf in oracle. In EDBT, pages 762–773, 2014

    work page 2014

  8. [8]

    R2g: a tool for migrating relations to graphs

    Roberto De Virgilio, Antonio Maccioni, and Riccardo Torlone. R2g: a tool for migrating relations to graphs. In EDBT, volume 2014, pages 640–643, 2014

    work page 2014

Show all 38 references
  1. [9]

    Software water- marking: Progress and challenges

    Ayan Dey, Sukriti Bhattacharya, and Nabendu Chaki. Software water- marking: Progress and challenges. INAE Letters, 4:65–75, 2019

    2019

  2. [10]

    Spearman’s footrule as a measure of disarray

    Persi Diaconis and Ronald L Graham. Spearman’s footrule as a measure of disarray. Journal of the Royal Statistical Society: Series B (Methodological), 39(2):262–268, 1977

    1977

  3. [11]

    Digital watermarking in the fractional fourier transformation domain

    Igor Djurovic, Srdjan Stankovic, and Ioannis Pitas. Digital watermarking in the fractional fourier transformation domain. Journal of Network and Computer Applications, 24(2):167–173, 2001

    2001

  4. [12]

    A new wavelet-based scheme for watermarking images

    Rakesh Dugad, Krishna Ratakonda, and Narendra Ahuja. A new wavelet-based scheme for watermarking images. In Proceedings 1998 International Conference on Image Processing. ICIP98 (Cat. No.98CB36269), volume 2, pages 419–423 vol.2, 1998

    1998

  5. [13]

    Goodrich, Jenny Lam, Nil Mamano, Michael Mitzenmacher, and Manuel Torres

    David Eppstein, Michael T. Goodrich, Jenny Lam, Nil Mamano, Michael Mitzenmacher, and Manuel Torres. Models and algorithms for graph watermarking. In Information Security, pages 283–301, 2016

    2016

  6. [14]

    Jean-Daniel Fekete. Reorder. js: A javascript library to reorder tables and networks. In IEEE VIS 2015 , 2015

    2015

  7. [15]

    Introduction to discrete transforms

    Jeffrey Fessler. Introduction to discrete transforms. Digital Signal Processing and Analysis , 3, 2005

    2005

  8. [16]

    Exploring network structure, dynamics, and function using networkx

    Aric Hagberg, Pieter Swart, and Daniel S Chult. Exploring network structure, dynamics, and function using networkx. Technical report, Los Alamos National Lab.(LANL), Los Alamos, NM (United States), 2008

    2008

  9. [17]

    Dissimilarity plots: A visual ex- ploration tool for partitional clustering

    Michael Hahsler and Kurt Hornik. Dissimilarity plots: A visual ex- ploration tool for partitional clustering. Journal of Computational and Graphical Statistics, 20(2):335–354, 2011

    2011

  10. [18]

    Multimedia watermarking techniques

    Frank Hartung and Martin Kutter. Multimedia watermarking techniques. Proceedings of the IEEE , 87(7):1079–1107, 1999

    1999

  11. [19]

    rotation, scale and translation invariant digital image watermarking

    JK Joseph. O’ruanaidh, and thierry. pun,“rotation, scale and translation invariant digital image watermarking”. Signal Processing , 66(3):303– 317, 1998

    1998

  12. [20]

    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. arXiv preprint arXiv:2301.10226 , 2023

    2023 arXiv

  13. [21]

    A recent survey on multimedia and database watermarking

    Sanjay Kumar, Binod Kumar Singh, and Mohit Yadav. A recent survey on multimedia and database watermarking. Multimedia Tools and Applications, 79:20149–20197, 2020

    2020

  14. [22]

    The konect project

    J ´erˆome Kunegis. The konect project. http://konect.cc/, 2013

    2013

  15. [23]

    Adversarial frontier stitching for remote neural network watermarking

    Erwan Le Merrer, Patrick Perez, and Gilles Tr ´edan. Adversarial frontier stitching for remote neural network watermarking. Neural Computing and Applications, 32:9233–9244, 2020

    2020

  16. [24]

    SNAP Datasets: Stanford large network dataset collection

    Jure Leskovec and Andrej Krevl. SNAP Datasets: Stanford large network dataset collection. http://snap.stanford.edu/data, June 2014

    2014

  17. [25]

    Graph sparsification with graph convolutional networks

    Jiayu Li, Tianyun Zhang, Hao Tian, Shengmin Jin, Makan Fardad, and Reza Zafarani. Graph sparsification with graph convolutional networks. International Journal of Data Science and Analytics , pages 1–14, 2022

    2022

  18. [26]

    Just how dense are dense graphs in the real world? a methodological note

    Guy Melancon. Just how dense are dense graphs in the real world? a methodological note. In Proceedings of the 2006 AVI workshop on BEyond time and errors: novel evaluation methods for information visualization, pages 1–7, 2006

    2006

  19. [27]

    A threshold selection method from gray-level his- tograms

    Nobuyuki Otsu. A threshold selection method from gray-level his- tograms. IEEE transactions on systems, man, and cybernetics , 9(1):62– 66, 1979

    1979

  20. [28]

    Template based recovery of fourier-based watermarks using log-polar and log-log maps

    Shelby Pereira, Joseph JK O Ruanaidh, Frederic Deguillaume, Gabriela Csurka, and Thierry Pun. Template based recovery of fourier-based watermarks using log-polar and log-log maps. In Proceedings IEEE International Conference on Multimedia Computing and Systems , vol- ume 1, pa...

    1999

  21. [29]

    Analysis of the gcr communication channel for image steganography

    Ante Poljicak, Davor Donevski, Petar Branislav Jelusic, Tamara Tomasegovic, and Tomislav Cigula. Analysis of the gcr communication channel for image steganography. In 2019 International Symposium ELMAR, pages 65–68, 2019

    2019

  22. [30]

    Graph classification with graph neural networks, 2024

    Pytorch. Graph classification with graph neural networks, 2024. Revi- sion 93aef5ee

    2024

  23. [31]

    Invisible watermarking schemes in spatial and frequency domains

    Saba Riaz, Muhammad Younus Javed, and Almas Anjum. Invisible watermarking schemes in spatial and frequency domains. In 2008 4th International Conference on Emerging Technologies , pages 211–216. IEEE, 2008

    2008

  24. [32]

    Rossi and Nesreen K

    Ryan A. Rossi and Nesreen K. Ahmed. The network data repository with interactive graph analytics and visualization. In AAAI, 2015

    2015

  25. [33]

    Image steganography using discrete fractional fourier transform

    Ashish Soni, Jitendra Jain, and Rakesh Roshan. Image steganography using discrete fractional fourier transform. In 2013 International Conference on Intelligent Systems and Signal Processing (ISSP) , pages 97–100. IEEE, 2013

    2013

  26. [34]

    Robust image watermarking via perceptual structural regularity-based jnd model

    Chunxing Wang, Meiling Xu, Wenbo Wan, Jian Wang, Lili Meng, Jing Li, and Jiande Sun. Robust image watermarking via perceptual structural regularity-based jnd model. KSII Transactions on Internet and Information Systems (TIIS) , 13(2):1080–1099, 2019

    2019

  27. [35]

    Wolfgang and Edward Delp

    Raymond B. Wolfgang and Edward Delp. A watermark for digital images. In Proceedings of 3rd IEEE International Conference on Image Processing, volume 3, pages 219–222 vol.3, 1996

    1996

  28. [36]

    Embedding robust labels into images for copyright protection

    Jian Zhao and Eckhard Koch. Embedding robust labels into images for copyright protection. In Proc. of the Int. Congress on Intellectual Property Rights for Specialized Information, Knowledge and New Tech- nologies, 1995

    1995

  29. [37]

    Watermarking graph neural networks by random graphs

    Xiangyu Zhao, Hanzhou Wu, and Xinpeng Zhang. Watermarking graph neural networks by random graphs. In 2021 9th International Symposium on Digital Forensics and Security (ISDFS) , pages 1–6. IEEE, 2021

    2021

  30. [38]

    Xiaohan Zhao, Qingyun Liu, Haitao Zheng, and Ben Y . Zhao. Towards graph watermarks. In Proceedings of the 2015 ACM on Conference on Online Social Networks , COSN ’15, page 101–112, New York, NY , USA, 2015. Association for Computing Machinery

    2015

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.