Writing study: Mechanism-Aware Ensemble Conditioning for Data-Limited Emulation of Extreme Events

NeurIPS overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← NeurIPS 2026 oral writing corpus

Isabella Thiel, Juan M. Bello-Rivas, Yannis Kevrekidis, Themis Sapsis · 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

42 physical pages

Before references / appendix

≈ 9.34 page-equivalents

Main figures / tables

5 / 3

Attached appendix figures / tables

6 / 10

Attached appendix span

≈ 23.9 page-equivalents

Before references / appendix: ≈ 22.24%References / intervening material: ≈ 20.86%Attached appendix: ≈ 56.9%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Source-bound correction: Top-level heading and reference/appendix boundaries checked against source PDF text. Bibliography titles/author initials, numbered contribution/property bullets, diagram steps, equations and TOC entries are not substantive headings. Section spans are page-coordinate estimates, not ink area or prose-only length. This is an author manuscript mirror, not a verified accepted-version template. Source layout uses physical single-column y/page-height, including equation-heavy pages.

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	≈ 1.54 pp.	3.67%
2 Method: Non-intrusive ensemble conditioning	3	≈ 3.02 pp.	7.19%
3 Low-dimensional example	6	≈ 0.91 pp.	2.17%
4 Experiments on high-dimensional chaotic PDE (QG)	7	≈ 1.56 pp.	3.71%
5 Limitations	8	≈ 1.29 pp.	3.07%
6 Conclusion	10	≈ 0.24 pp.	0.57%
References	10	≈ 1.86 pp.	4.43%
A Proofs of main results	12	≈ 3.9 pp.	9.29%
B Low-Dimensional System: methodology and implementation details	16	≈ 6 pp.	14.29%
C QG model: governing equations and numerical implementation	22	≈ 2.53 pp.	6.02%
D QG model: methodology and implementation details	24	≈ 5.11 pp.	12.17%
E QG model: evaluation and additional results	29	≈ 6.36 pp.	15.14%
NeurIPS Paper Checklist	36	≈ 6.91 pp.	16.45%

All detected figure and table locations

Main text

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

Attached appendix

• Figure 6 · p. 16
• Table 4 · p. 19
• Table 5 · p. 20
• Table 6 · p. 21
• Table 7 · p. 21
• Table 8 · p. 22
• Figure 7 · p. 23
• Figure 8 · p. 28
• Table 9 · p. 29
• Table 10 · p. 31
• Table 11 · p. 32
• Table 12 · p. 32
• Table 13 · p. 33
• Figure 9 · p. 33
• Figure 10 · p. 34
• Figure 11 · p. 35

Argument and organization

1. 1 Introduction

   Links scarce extreme events to a missing conditioning signal.

2. 2 Method: Non-intrusive ensemble conditioning

   Derives a covariance-based instability proxy and conditioning interface.

3. 3 Low-dimensional example

   Tests the geometric mechanism in a controlled low-dimensional system.

4. 4 Experiments on high-dimensional chaotic PDE (QG)

   Evaluates rare-event diagnostics in a physical PDE emulator.

5. 5 Limitations

   States when the proxy and conditioning benefit are limited.

6. 6 Conclusion

   Connects the mechanism back to data-limited emulation.

Problem, claim and contribution

The paper motivates the representation through local instability in rare physical events: global average prediction error is not sufficient evidence that an emulator preserves the event geometry. This frames the conditioning signal and the tail-sensitive evaluation together.

PDF pp.1–3, Introduction and contribution list · Main paper

Argument and section order

Nudged ensemble dynamics lead to covariance geometry, finite-ensemble guarantees, and a FiLM conditioning interface before the low-dimensional and quasi-geostrophic experiments. Theoretical statements explain the proxy rather than replacing an empirical application test.

PDF pp.3–7, Sections 2–4; Theorems 2.1–2.2 · Main paper

Figures, captions and visual explanation

Figure 1 distinguishes the reference trajectory, nudged coarse trajectory, ensemble paths, and unstable subspace ellipsoids in one geometric scene. Figure 4 aligns parameter-sweep panels and normalizes both diagnostics to the same long-data baseline.

PDF p.3, Figure 1; p.8, Figure 4 · Main paper

Tables, comparisons and evidence

Table 2 identifies the common backbone, loss, optimizer, training setup, and the added interface, while retaining metrics where the long-data baseline wins. Table 1 is explicitly sweep-selected sensitivity evidence, not the main model-selection result; this distinction is essential to the comparison claim.

PDF p.7, Table 1; p.8, Table 2 · Main paper

Appendix and reproducibility

The appendices separate mathematical proofs, toy-system algorithms, the physical model, correction/training/evaluation algorithms, and extended threshold diagnostics. The supplementary runtime/memory plot asks how ensemble size changes cost rather than treating conditioning as free.

PDF pp.12–35, Appendices A–E; p.28, Figure 8 · Main paper

Clarity and writing conventions

The manuscript distinguishes an instability probe from uncertainty quantification and follows the main comparison with limits on when coarse/reference mismatch supplies useful information. Its strongest wording is tied to threshold exceedance and event-frequency diagnostics, not every metric.

PDF p.3, Section 2.1; pp.8–9, Section 5 · Main paper

What to reuse—and what not to infer

Transferable lesson: For scientific ML, connect the physical mechanism to an explicit measured diagnostic and preserve both matched controls and unfavorable metrics.

Boundary: Ensemble covariance has stated small-noise/regularity and sampling assumptions; the source does not establish universal rare-event improvement. Sweep-selected supplementary configurations must remain distinct from fixed held-out comparisons. The public manuscript revision is not confirmed to match accepted OpenReview.

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 source scan of framing, argument roadmap, selected substantive passages, method/theorem statements, main figure/table captions, evidence and limits across PDF pp.1–10. Two selected main-body pages were actually inspected as rendered images. This is not exhaustive line-by-line technical verification or experiment reproduction. Title and all author names compared against official program metadata and manuscript first page; matching authors, with capitalization, middle initials or expanded given names where applicable.

Appendix / supplement

Appendix proof/setup openings, algorithm headings, result-table captions, and runtime/memory figure caption were inspected for evidence organization; no claim of complete proof verification.

Visually inspected PDF pages

3, 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 manuscript arXiv:2609.30746v1; not verified identical to the accepted OpenReview revision · 42 pages