Writing study: Non-Colliding Biometric Identities for Digital Entities: Geometry, Capacity, and Million-Scale Virtual Identity Provisioning

NeurIPS overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← NeurIPS 2026 oral writing corpus

Yuyang Ji, Yixuan Shen, Anil Jain, Xiaoming Liu, Feng Liu · Official NeurIPS 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: Domain-grounded methods & physical 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

25 physical pages

Before references / appendix

≈ 9.09 page-equivalents

Main figures / tables

5 / 3

Attached appendix figures / tables

6 / 1

Attached appendix span

≈ 12.91 page-equivalents

Before references / appendix: ≈ 36.36%References / intervening material: ≈ 12%Attached appendix: ≈ 51.64%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	≈ 1.49 pp.	5.96%
2 Related Work	3	≈ 0.55 pp.	2.2%
3 Methodology	3	≈ 3.82 pp.	15.28%
4 Experiments	7	≈ 2.06 pp.	8.24%
5 Conclusion	9	≈ 0.43 pp.	1.72%
References	10	≈ 3 pp.	12%
Appendix	13	≈ 0.44 pp.	1.76%
A Broader Impact and Ethics	13	≈ 0.13 pp.	0.52%
B Verification Threshold Selection	13	≈ 0.84 pp.	3.36%
C Capacity Bound Derivation	14	≈ 7.18 pp.	28.72%
D Additional Implementation Details of G (GapGen)	21	≈ 0.76 pp.	3.04%
E Verification, Visual Examples, and IAPCT	22	≈ 2.13 pp.	8.52%
F Additional Results	24	≈ 1.43 pp.	5.72%

All detected figure and table locations

Main text

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

Attached appendix

• Figure 6 · p. 14
• Figure 7 · p. 15
• Table 4 · p. 18
• Figure 8 · p. 22
• Figure 9 · p. 23
• Figure 10 · p. 24
• Figure 11 · p. 25

Argument and organization

1. Problem and contribution

   The introduction treats virtual identity provisioning as a different problem from generating a plausible face. The formal problem makes three requirements inspectable: separation from the real gallery, separation between virtual identities, and image realizability. Remark 1 then restricts the guarantee to the gallery and provisioning time actually considered.

2. Argument sequence

   The argument follows the provisioned object through stages: define identity-space requirements, allocate embeddings geometrically, synthesize images with GapGen, then evaluate verification and virtual/real discrimination. The appendix separates threshold calibration and capacity arguments from implementation and additional tests.

3. Evidence and verification

   The results distinguish an allocated embedding satisfying geometric criteria from a rendered face still satisfying them after re-encoding. Table 1 varies scale and thresholds; Table 2 includes image resolution, image quality and round-trip behavior; Table 3 separates detection and verification protocols. This stage-specific comparison is more informative than a single success percentage.

Problem, claim and contribution

The introduction treats virtual identity provisioning as a different problem from generating a plausible face. The formal problem makes three requirements inspectable: separation from the real gallery, separation between virtual identities, and image realizability. Remark 1 then restricts the guarantee to the gallery and provisioning time actually considered.

Introduction pp. 1–2; §3.1 and Remark 1, pp. 3–4 · Main paper

Argument and section order

The argument follows the provisioned object through stages: define identity-space requirements, allocate embeddings geometrically, synthesize images with GapGen, then evaluate verification and virtual/real discrimination. The appendix separates threshold calibration and capacity arguments from implementation and additional tests.

§§3.1–3.4, pp. 3–7; §§4–5, pp. 7–9; Appendices B–F, pp. 13–25 · Main paper

Figures, captions and visual explanation

Figure 2 is a left-to-right provisioning diagram with visibly separated allocation, generation and evaluation regions. It marks frozen versus trainable components and connects each block to a method subsection. On p. 8, the paper places an embedding-space scalability table separately from an image-space round-trip table: the layout makes the change of measured object visible.

Figure 2, p. 4; Tables 1–2 and Figure 4, p. 8; rendered pages 4 and 8 · Main paper

Tables, comparisons and evidence

The results distinguish an allocated embedding satisfying geometric criteria from a rendered face still satisfying them after re-encoding. Table 1 varies scale and thresholds; Table 2 includes image resolution, image quality and round-trip behavior; Table 3 separates detection and verification protocols. This stage-specific comparison is more informative than a single success percentage.

Tables 1–2, p. 8; Table 3, p. 9; Appendix E, pp. 22–24 · Main paper

Appendix and reproducibility

The appendix supplies threshold curves, capacity derivations, model configuration and evaluation definitions. The ambient-dimensional capacity argument and the empirical manifold discussion are different kinds of evidence and should remain labeled as such. Ethics and identity-update limitations are not buried solely in implementation detail.

Appendices A–C, pp. 13–20; D–F, pp. 21–25 · Main paper

Clarity and writing conventions

The strongest reusable wording is the operational definition of each requirement and the explicit gallery-relative qualification. Broad phrases about non-collision need that qualification nearby. The Table 1 caption's absolute wording should not be copied without checking the individual cells, which include a less-than-100 separation entry.

§3.1 and Remark 1, pp. 3–4; Table 1 caption and cells, p. 8 · Main paper

What to reuse—and what not to infer

Transferable lesson: Define the objects and operating conditions first, draw the transformations between them, and report success separately before and after each transformation.

Boundary: The author manuscript does not establish a permanent open-world identity guarantee. Embedding separation, image round-trip success and virtual/real detection are distinct claims; the cited public revision is not verified identical to the accepted revision.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-10-02

Reading status

Writing-focused manuscript scan + selected visuals inspected

Main text

Writing-oriented scan across manuscript framing, section sequence, figure/table captions, selected argument/evaluation prose and conclusion; not exhaustive technical or proof verification. Sections 1–5, physical PDF pages 1–9; references pp. 10–12. Rendered Figure 2 and the stage-specific comparison tables on pp. 4 and 8. Source is the exact public author version listed below, not a verified accepted/camera-ready revision.

Appendix / supplement

Writing-oriented appendix/documentation scan: Appendices A–F, pp. 13–25: ethics, operating thresholds, geometric capacity analysis, GapGen implementation, verification protocols and additional results. Source code, data, model outputs and external video assets were not executed or independently validated.

Visually inspected PDF pages

4, 8

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.

• Public author arXiv manuscript 2605.18238v1; not verified identical to the accepted or camera-ready OpenReview revision · 25 pages