Writing study: PateGAIL++: Utility Optimized Private Trajectory Generation with Imitation Learning

ICLR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICLR 2026 oral writing corpus

Yingjie Ma, Bijal Bharadva, Xin Zhang, Joann Qiongna Chen · Official ICLR 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: Learning & optimization algorithms (primary) · Theory & formal foundations.

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

16 physical pages

Before references / appendix

≈ 10.11 page-equivalents

Main figures / tables

2 / 7

Attached appendix figures / tables

1 / 3

Attached appendix span

≈ 2.89 page-equivalents

Before references / appendix: ≈ 63.19%References / intervening material: ≈ 18.75%Attached appendix: ≈ 18.06%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

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	≈ 1.06 pp.	6.63%
2 PRELIMINARIES	2	≈ 1.8 pp.	11.25%
3 METHOD	4	≈ 2.79 pp.	17.44%
4 EVALUATION	7	≈ 3.37 pp.	21.06%
5 CONCLUSION	10	≈ 0.22 pp.	1.38%
6 ACKNOWLEDGMENT	10	≈ 0.24 pp.	1.5%
REFERENCES	11	≈ 3 pp.	18.75%
A APPENDIX	14	≈ 2.89 pp.	18.06%

All detected figure and table locations

Main text

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

Attached appendix

• Table 8 · p. 15
• Table 9 · p. 15
• Figure 3 · p. 15
• Table 10 · p. 16

Argument and organization

1. Privacy challenge to three design components

   Maps sensitivity, accounting and training stability to the proposed method.

2. Utility, attacks and deployment variants

   Separates statistical fidelity from empirical privacy attacks and local-noise experiments.

Problem, claim and contribution

The introduction motivates adaptive noise using distinctive versus common trajectories, making the application-level concern concrete before introducing DP notation.

Figure 1 and Section 1, pp. 1–2 · Main paper

Argument and section order

Three numbered challenges are explicitly mapped to method subsections. This is a reusable signposting device, although the promised formal accounting proof is not actually supplied in the inspected paper.

Section 2.4–3, p. 4; Section 3.2, pp. 5–6 · Main paper

Figures, captions and visual explanation

The framework figure separates user devices from the server and places reward equations inside the server box. A fuller caption should distinguish this central aggregation view from the later local-privacy variant.

Figure 2, p. 4 · Main paper

Tables, comparisons and evidence

Attack tables display both accuracy and AUC with variation; however, the stated 70/30 membership imbalance makes a naive 50% accuracy chance baseline misleading.

Section 3.4, p. 7; Tables 4–5, p. 9 · Main paper

Appendix and reproducibility

Additional attacks and participation results are findable by subsection, but generic reproducibility assurances are stronger than the concrete architecture, clipping and privacy-accounting details documented here.

Appendix A.1–A.4, pp. 14–16 · Main paper

Clarity and writing conventions

The prose explicitly warns that centralized baselines are not an apples-to-apples comparison, a useful practice. Later blanket superiority claims should instead acknowledge the mixed rows.

Section 4.1, pp. 7–8; Tables 3, 5, 7 · Main paper

What to reuse—and what not to infer

Transferable lesson: Use challenge-to-component signposting and separate privacy threat models visibly; qualify the narrative against every metric rather than converting mixed evidence into a uniform success story.

Boundary: This is also a cautionary writing study. Table 1 is noise-free despite surrounding DP framing; sensitivity bounds, clipped critic range and composition including the pilot need clarification. LiRA AUC does not consistently improve, and selecting the best hyperparameter separately for each metric is not one deployable configuration.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-09-19

Reading status

Full text + selected visuals inspected

Main text

Pages 1–10, Sections 1–6 including privacy definitions, adaptive noise, central/local settings and experiments.

Appendix / supplement

Pages 14–16, Appendices A.1–A.4 including ethics, baseline descriptions, extra attacks and participation ablations.

Visually inspected PDF pages

4, 9

• Official ICLR 2026 proceedings conference PDF; PDF creation metadata: Thu Mar 5 13:10:04 2026 EST · 16 pages