Writing study: MV-FGAD: Towards Efficient and Effective Federated Graph Anomaly Detection via Multi-view Learning

ICML overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICML 2026 oral writing corpus

Junyi Yan, Ke Liang, Hao Yu, Meng Liu, Hao Tan, Tianrui Liu, Jun-Jie Huang, Xinwang Liu · Official ICML 2026 oral record

Observed writing choices, not acceptance rules. This study examines the inspected manuscript version. It does not establish why the paper was accepted or that its presentation choices should be copied unchanged.

Figure, table and section ledger

Editorial contribution type: Efficiency, inference & systems (primary).

Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.

Canonical PDF

20 physical pages

Before references / appendix

≈ 8.62 page-equivalents

Main figures / tables

10 / 3

Attached appendix figures / tables

5 / 4

Attached appendix span

≈ 7.91 page-equivalents

Before references / appendix: ≈ 43.1%References / intervening material: ≈ 17.35%Attached appendix: ≈ 39.55%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Source-bound correction: Table 3 on physical p.7 is split into separate Table and 3. caption fragments. The source image and glyph positions confirm this main-body scorer-ablation table.

Section-by-section lengths and source pages

The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.

Top-level heading	Starts on PDF page	Estimated span	Share of entire PDF
1. Introduction	1	≈ 1.44 pp.	7.2%
2. Related Work	2	≈ 1.06 pp.	5.3%
3. Problem Statement	3	≈ 0.42 pp.	2.1%
4. Methodology	4	≈ 2.25 pp.	11.25%
5. Experiments	6	≈ 2.71 pp.	13.55%
6. Conclusion	9	≈ 0.16 pp.	0.8%
Acknowledgement	9	≈ 0.16 pp.	0.8%
Impact Statement	9	≈ 0.05 pp.	0.25%
References	9	≈ 3.46 pp.	17.3%
A. Algorithm and Complexity	13	≈ 0.77 pp.	3.85%
B. Detailed Description of Datasets	13	≈ 1 pp.	5%
C. Detailed Description of Baselines	14	≈ 0.9 pp.	4.5%
D. Details of Implementation	15	≈ 0.72 pp.	3.6%
E. Additional Experimental Results	16	≈ 4.53 pp.	22.65%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 1
• Figure 3 · p. 2
• Figure 4 · p. 3
• Figure 5 · p. 5
• Table 1 · p. 7
• Figure 6 · p. 7
• Figure 7 · p. 7
• Table 2 · p. 7
• Table 3 · p. 7
• Figure 8 · p. 8
• Figure 9 · p. 8
• Figure 10 · p. 9

Attached appendix

• Table 4 · p. 14
• Table 5 · p. 16
• Table 6 · p. 17
• Table 7 · p. 17
• Figure 11 · p. 17
• Figure 12 · p. 18
• Figure 13 · p. 18
• Figure 14 · p. 19
• Figure 15 · p. 19

Argument and organization

1. 1 Introduction; 2 Related Work; 3 Problem Statement

   Separates partitioned-neighborhood limitations from weak-anomaly detection and defines client subgraphs.

2. 4 Methodology

   Moves from shared federated knowledge to three local views and a common Mahalanobis scorer.

3. 5 Experiments; 6 Conclusion

   Separates detection performance, efficiency, view/scorer ablations and client-level score-distribution examples.

Problem, claim and contribution

The introduction names two different bottlenecks: missing cross-client neighborhood information and weak anomalous signals.

Sections 1 and 3, physical pp.1–4 · Main paper

Argument and section order

Method subsections separate knowledge transfer, representations and scoring, making the views distinct from the downstream score rule.

Sections 4.1–4.3, pp.4–6 · Main paper

Figures, captions and visual explanation

The framework follows server communication into three color-coded local branches and a shared scorer; the result page juxtaposes metrics, runtime, communication and two ablation tables.

Figure 4, p.3; Tables 1–3 and Figures 6–7, p.7 · Main paper

Tables, comparisons and evidence

AUROC and AUPRC are reported together, but the published average rank additionally penalizes large standard deviations by assigning last place. This is not an ordinary rank by mean metric.

Table 1 caption, p.7; stability discussion, p.8 · Main paper

Appendix and reproducibility

Full seven-dataset scorer and aggregation tables reveal exceptions omitted from the main three-dataset scorer table; implementation reports three trials and hardware.

Appendix D–E.2, pp.15–17; Tables 5–7 · Main paper

Clarity and writing conventions

The communication claim separates parameter size from the number of rounds needed to achieve performance, rather than treating a small model as sufficient evidence of low communication cost.

Section 5.3, p.8; Appendix E.2 and Figure 11, p.17 · Main paper

What to reuse—and what not to infer

Transferable lesson: For an efficient detector, show quality, runtime, communication and component contributions as separate evidence, and define any nonstandard ranking rule in the caption.

Boundary: Do not interpret the adjusted rank as a standard leaderboard rank or the strongest mean as a universal win. In Appendix Table 6, Mahalanobis scoring loses to alternatives on Elliptic and on Questions AUPRC. Client partitioning and parameter sharing do not themselves establish a privacy guarantee; a single client’s score-density plot is not population-wide mechanism validation.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-10-07

Reading status

Writing-focused manuscript scan + selected visuals inspected

Main text

Writing-focused scan across physical PDF pp.1–9, covering framing, section progression, definitions, method/evidence explanations, captions and scope. Physical pages extracted individually to avoid form-feed pagination errors. Selected full-page rasters inspected; not exhaustive line-by-line technical verification or experimental reproduction.

Appendix / supplement

Appendix A–E structure inventoried; selected reading of baseline adaptations and implementation D (pp.15–16), complete view/scorer/aggregation ablations and communication analysis E.1–E.2 (pp.16–17). All extra case plots and software behavior were not independently reproduced.

Visually inspected PDF pages

3, 7

This is a writing-focused scan of the manuscript and the stated supplementary scope: framing, organization, visual explanation, evidence presentation and wording. It is not a line-by-line proof audit, reproduction of results or exhaustive scientific review.

• Published proceedings PDF, PMLR volume 306, retrieved 2026-10-07 · 20 pages