Writing study: Scalable Event Cloud Network for Event-based Classification

ICML overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICML 2026 oral writing corpus

Hongwei Ren, Fei Ma, Xiaopeng Lin, Yuetong Fang, Hongxiang Huang, Yue Zhou, Yulong Huang, Haotian Fu, Ziyi Yang, Youxin Jiang, Xiangqian Wu, Bojun Cheng · 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: Architecture & representation (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

18 physical pages

Before references / appendix

≈ 9.04 page-equivalents

Main figures / tables

4 / 11

Attached appendix figures / tables

4 / 2

Attached appendix span

≈ 5.77 page-equivalents

Before references / appendix: ≈ 50.22%References / intervening material: ≈ 17.72%Attached appendix: ≈ 32.06%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Source-bound correction: Replace a prose cross-reference falsely detected as a caption with the actual caption from physical-page source text and coordinates.

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	≈ 1.5 pp.	8.33%
2. Related Work	2	≈ 1.21 pp.	6.72%
3. Method	4	≈ 2.76 pp.	15.33%
4. Experiment	6	≈ 2.82 pp.	15.67%
5. Conclusion	9	≈ 0.07 pp.	0.39%
Acknowledgements	9	≈ 0.11 pp.	0.61%
Impact Statement	9	≈ 0.21 pp.	1.17%
References	10	≈ 3.19 pp.	17.72%
A. References to comparative method abbreviations	13	≈ 0.21 pp.	1.17%
B. Comparison of Event Cloud and Voxel Representations	13	≈ 0.36 pp.	2%
C. Implement Details	13	≈ 0.71 pp.	3.94%
D. Dataset	14	≈ 0.15 pp.	0.83%
E. Super-linear Increase	14	≈ 0.12 pp.	0.67%
F. Specific Dimension Variance in SECNet	14	≈ 0.68 pp.	3.78%
G. Hardware Resource Performance	15	≈ 0.15 pp.	0.83%
H. Pseudo Code	15	≈ 0.98 pp.	5.44%
I. Ablation Study on Spatial-FA Module	16	≈ 0.17 pp.	0.94%
J. Look up Table for Runtime	16	≈ 1.46 pp.	8.11%
K. Spatio-temporal Features	18	≈ 0.15 pp.	0.83%
L. Visualization	18	≈ 0.63 pp.	3.5%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 2
• Figure 3 · p. 3
• Table 1 · p. 6
• Figure 4 · p. 6
• Table 2 · p. 7
• Table 3 · p. 7
• Table 4 · p. 7
• Table 5 · p. 8
• Table 6 · p. 8
• Table 7 · p. 8
• Table 8 · p. 8
• Table 9 · p. 9
• Table 10 · p. 9
• Table 11 · p. 13

Attached appendix

• Table 12 · p. 14
• Figure 5 · p. 14
• Figure 6 · p. 16
• Figure 7 · p. 16
• Figure 8 · p. 17
• Table 13 · p. 18

Argument and organization

1. 1 Introduction; 2 Related Work

   Motivates representation conversion costs and scalability limits in event processing.

2. 3 Method

   Defines event-cloud structure, grouping/sampling and spatial/temporal frequency-aware modules.

3. 4 Experiments; 5 Conclusion

   Separates preprocessing, task accuracy, event-count scaling and hardware implementation evidence.

Problem, claim and contribution

The opening treats preprocessing and loss of fine temporal structure as part of the modeling problem rather than presenting accuracy alone as the objective.

Abstract and Introduction, p.1 · Main paper

Argument and section order

The experiments separate object recognition, action recognition and pose estimation, then test module removal, event count and FPGA implementation.

Sections 4.2-4.7, pp.7-9 · Main paper

Figures, captions and visual explanation

The architecture overview combines a full hierarchy with enlarged grouping, frequency, aggregation and residual modules, using repeated symbols and tensor dimensions to connect scales of explanation.

Figure 3, p.3 · Main paper

Tables, comparisons and evidence

The classification table separates ImageNet-pretrained comparators from models without pretraining; hardware tables keep accuracy, parameters, operations and speed visible together.

Table 2, p.7; Tables 6-8, p.8 · Main paper

Appendix and reproducibility

The supplement explicitly acknowledges that the convolution replacement can score better on most datasets while costing more, and ties throughput to input-event/centroid configurations.

Appendices I-J, p.16; pipeline pseudocode, p.15 · Main paper

Clarity and writing conventions

The main conclusion uses strong evaluative language, but the most reusable claim is the narrower accuracy-compute trade-off documented by the supplement and the named hardware conditions.

Conclusion and FPGA comparison, p.9; Appendix I, p.16 · Main paper

What to reuse—and what not to infer

Transferable lesson: For efficient sensing architectures, distinguish conversion cost, network cost and hardware measurements, and expose the accuracy cost of the efficient module.

Boundary: Module-level FPGA measurements are not full-system energy validation; throughput and accuracy depend on event count, centroids, pretraining and task configuration.

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-oriented scan of the main argument through physical p.9, with focused physical-page text inspection on 1, 3, 5, 7, 9. Framing, section progression, mechanism/proof explanation, displays, comparison conditions and conclusions were inspected. This is not exhaustive technical reading, proof verification or reproduction.

Appendix / supplement

Appendix organization and main-text pointers were checked; focused text inspection of physical pages 13, 15, 16. Remaining appendix proofs and details were not exhaustively read. No separately linked supplement inspected.

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.

• Published proceedings PDF, PMLR volume 306, retrieved 2026-10-07 · 18 pages