Writing study: LASER: Learning Active Sensing for Continuum Field Reconstruction

ICML overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICML 2026 oral writing corpus

Huayu Deng, Jinghui Zhong, Xiangming Zhu, Yunbo Wang, Xiaokang Yang · 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: 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

26 physical pages

Before references / appendix

≈ 9.28 page-equivalents

Main figures / tables

6 / 11

Attached appendix figures / tables

12 / 3

Attached appendix span

≈ 13.91 page-equivalents

Before references / appendix: ≈ 35.69%References / intervening material: ≈ 10.81%Attached appendix: ≈ 53.5%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	≈ 1.08 pp.	4.15%
2. Related Work	2	≈ 0.39 pp.	1.5%
3. Problem Definition	2	≈ 0.39 pp.	1.5%
4. Method	3	≈ 2.88 pp.	11.08%
5. Experiments	6	≈ 3.8 pp.	14.62%
6. Conclusions and Limitations	9	≈ 0.2 pp.	0.77%
Impact Statement	10	≈ 0.15 pp.	0.58%
Acknowledgements	10	≈ 0.08 pp.	0.31%
References	10	≈ 2.81 pp.	10.81%
A. Algorithm	13	≈ 0.61 pp.	2.35%
B. Continuum Field Latent World Model	13	≈ 1.97 pp.	7.58%
C. Benchmarks	15	≈ 0.42 pp.	1.62%
D. Compared Methods	16	≈ 0.26 pp.	1%
E. Implementation Details	16	≈ 1.92 pp.	7.38%
F. Additional Results	18	≈ 8.73 pp.	33.58%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Table 1 · p. 2
• Figure 2 · p. 3
• Figure 3 · p. 4
• Figure 4 · p. 4
• Table 2 · p. 6
• Figure 5 · p. 6
• Table 3 · p. 7
• Table 4 · p. 7
• Figure 6 · p. 8
• Table 5 · p. 8
• Table 6 · p. 8
• Table 7 · p. 8
• Table 8 · p. 9
• Table 9 · p. 9
• Table 10 · p. 9
• Table 11 · p. 9

Attached appendix

• Table 12 · p. 16
• Table 13 · p. 17
• Table 14 · p. 18
• Figure 7 · p. 19
• Figure 8 · p. 19
• Figure 9 · p. 20
• Figure 10 · p. 21
• Figure 11 · p. 22
• Figure 12 · p. 23
• Figure 13 · p. 24
• Figure 14 · p. 24
• Figure 15 · p. 25
• Figure 16 · p. 25
• Figure 17 · p. 26
• Figure 18 · p. 26

Argument and organization

1. Problem and contribution

   Frames active sensing as anticipating informative future locations, while stating that sensor observations are queried from offline high-fidelity data rather than live physical interaction.

2. Argument sequence

   The POMDP names observations, latent states, actions and rewards before separating the world model, policy and RL algorithm; experimental analyses return to these components.

3. Evidence and verification

   World-model shift, reward horizon and heuristic policies receive separate controls; initialization variability is measured over three sampled sensor placements, not three independently trained models.

Problem, claim and contribution

Frames active sensing as anticipating informative future locations, while stating that sensor observations are queried from offline high-fidelity data rather than live physical interaction.

Introduction and Problem Definition, pp. 1-2 · Main paper

Argument and section order

The POMDP names observations, latent states, actions and rewards before separating the world model, policy and RL algorithm; experimental analyses return to these components.

Sections 3-5, pp. 2-9 · Main paper

Figures, captions and visual explanation

The graphical models reuse encoder/dynamics/decoder colors across prediction and decision panels. Neighboring result tables distinguish forward rollout from reconstruction with new observations, and retain per-dataset unit multipliers.

Figure 2, p. 3; Tables 3-4, p. 7; both PNGs inspected · Main paper

Tables, comparisons and evidence

World-model shift, reward horizon and heuristic policies receive separate controls; initialization variability is measured over three sampled sensor placements, not three independently trained models.

Tables 8-11, p. 9; Table 14 and Appendix F.1, p. 18 · Main paper

Appendix and reproducibility

Architecture and training settings precede sensitivity tests; extended figures show rollout degradation and how observation budgets affect active sensing.

Appendices B and E-F, pp. 13-26 · Main paper

Clarity and writing conventions

The conclusion names surrogate-model fidelity and policy overhead as limitations, though some result prose is stronger than individual table cells justify.

Section 6, p. 9; Table 4, p. 7 · Main paper

What to reuse—and what not to infer

Transferable lesson: Give distinct protocols distinct result tables, especially when initial-observation prediction and continually observed reconstruction use similar error notation.

Boundary: Archived real-world SST data do not demonstrate deployed physical sensors. Preserve non-winning cells and do not relabel placement variability as training-seed uncertainty.

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 distributed manuscript scan, not exhaustive technical/proof validation. Sections 1-6, pp. 1-9: sensing motivation, latent POMDP/world-model definition, policy learning, PDE and SST comparisons, controlled analyses and limitations.

Appendix / supplement

Writing-focused appendix scan; code, data and external assets not executed. Appendices A-F, pp. 13-26: algorithm, encoder/dynamics/decoder, datasets and baselines, training setup, initialization controls, rollout and active-sensing visualizations.

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 ICML 2026 proceedings PDF, PMLR volume 306; retrieved 2026-10-07. Public OpenReview decision notes were not independently reverified. · 26 pages