Writing study: ConFlux: Multivariate Time Series in Flux, One Unified Forecast in Confluence

ICML overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICML 2026 oral writing corpus

Shiyu Wang, Juntong Ni, Ziyi Zhang, Baichuan Mo, Xinyue Zhong, Chengxin Wang, Yuchen Fang, Zhou Ye, Yang Xiang · 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

31 physical pages

Before references / appendix

≈ 8.1 page-equivalents

Main figures / tables

6 / 4

Attached appendix figures / tables

20 / 9

Attached appendix span

≈ 19.91 page-equivalents

Before references / appendix: ≈ 26.13%References / intervening material: ≈ 9.65%Attached appendix: ≈ 64.23%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	≈ 1.46 pp.	4.71%
2. Related Work	2	≈ 0.73 pp.	2.35%
3. Problem Definition	3	≈ 0.22 pp.	0.71%
4. Methodology	3	≈ 2.5 pp.	8.06%
5. Experiment	6	≈ 2.55 pp.	8.23%
6. Conlusion	8	≈ 0.22 pp.	0.71%
Impact Statement	9	≈ 0.05 pp.	0.16%
References	9	≈ 2.99 pp.	9.65%
A. Details of Five Complementary Association Measures	12	≈ 1 pp.	3.23%
B. Theoretical Analysis of Variate Sorting	13	≈ 2.05 pp.	6.61%
C. DFT as Cosine/Sine Basis Coefficients	15	≈ 0.95 pp.	3.06%
D. Evaluation Datasets.	16	≈ 0.18 pp.	0.58%
E. Baseline Details	16	≈ 1.31 pp.	4.23%
F. Metrics	17	≈ 0.17 pp.	0.55%
G. Implementation Details	17	≈ 0.92 pp.	2.97%
H. Forecasting with From-scratch Training.	18	≈ 0.18 pp.	0.58%
I. Large-Scale Dimensional Experiment.	18	≈ 0.8 pp.	2.58%
J. Patch Size Impact	19	≈ 1.68 pp.	5.42%
K. Lookback Window Length Impact	21	≈ 0.27 pp.	0.87%
L. Inference Efficiency	21	≈ 0.95 pp.	3.06%
M. Forecasting without In-domain Fine-tuning — Full Results	22	≈ 0.29 pp.	0.94%
N. Forecasting with In-domain Fine-tuning — Full Results	22	≈ 0.41 pp.	1.32%
O. Forecasting with From-scratch Training — Full Results	23	≈ 0.23 pp.	0.74%
P. Broad Impact	23	≈ 2.96 pp.	9.55%
Q. Limitation and Future Work	26	≈ 0.63 pp.	2.03%
R. Additional Visualizations of Variate Sorting	27	≈ 2 pp.	6.45%
S. Show Case	29	≈ 2.91 pp.	9.39%

All detected figure and table locations

Main text

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

Attached appendix

• Table 5 · p. 17
• Table 6 · p. 18
• Table 7 · p. 18
• Table 8 · p. 19
• Figure 7 · p. 19
• Table 9 · p. 20
• Figure 8 · p. 21
• Table 10 · p. 22
• Table 11 · p. 24
• Table 12 · p. 25
• Table 13 · p. 26
• Figure 9 · p. 27
• Figure 10 · p. 27
• Figure 11 · p. 27
• Figure 12 · p. 28
• Figure 13 · p. 28
• Figure 14 · p. 28
• Figure 15 · p. 28
• Figure 16 · p. 29
• Figure 17 · p. 29
• Figure 18 · p. 29
• Figure 19 · p. 30
• Figure 20 · p. 30
• Figure 21 · p. 30
• Figure 22 · p. 30
• Figure 23 · p. 31
• Figure 24 · p. 31
• Figure 25 · p. 31
• Figure 26 · p. 31

Argument and organization

1. 1 Introduction; 2 Related Work; 3 Problem Definition

   Moves from a multistream analogy to variable-count and horizon notation.

2. 4 Methodology

   Defines association-based ordering, patch aggregation, unpatching and a Fourier decoder.

3. 5 Experiment; 6 Conclusion

   Separates pretraining data, transfer modes, ablations, scaling and efficiency analyses.

Problem, claim and contribution

The confluence metaphor motivates integration, but the technical claim is specific: sorting makes adjacent-variable patching useful and reduces the number of attention tokens.

Introduction, p.1; Sections 4.1-4.3, pp.4-5 · Main paper

Argument and section order

The method stages follow the same order as the architecture diagram, while training corpus and no-finetuning evaluation are separately identified.

Method overview, p.3; Sections 5.1-5.3, p.6 · Main paper

Figures, captions and visual explanation

A full-width diagram includes a graphical legend and discloses an omitted feed-forward block; matched before/after association matrices share a color scale beside scaling and efficiency curves.

Figure 3, p.4; Figures 4-6, p.8 · Main paper

Tables, comparisons and evidence

No-finetuning, in-domain finetuning and from-scratch training are different regimes; individual metric losses remain visible despite broad state-of-the-art wording.

Table 1, p.6; Tables 6-7, p.18 · Main paper

Appendix and reproducibility

The sorting argument proves a spectral relaxation and nearest-permutation projection, not unrestricted optimality of the original discrete ordering problem.

Appendix B, pp.13-14 · Main paper

Clarity and writing conventions

Efficiency reporting needs reconciliation: the plotted latency axis uses seconds while Appendix L describes microseconds, so the numerical speed scale cannot be confidently transferred.

Figure 5, p.8; Appendix L, p.21 · Main paper

What to reuse—and what not to infer

Transferable lesson: Align an architectural motivation with the exact reduction it achieves, and keep training regimes and measurement units consistent across main text and appendix.

Boundary: The sorting visualization is qualitative association evidence, not a causal explanation; complexity reduction to fewer patch tokens is not linear attention, and the runtime unit conflict remains unresolved.

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.8, with focused physical-page text inspection on 1, 3, 4, 5, 6, 7, 8, 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, 14, 18, 21. Remaining appendix proofs and details were not exhaustively read. No separately linked supplement inspected.

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.

• Published proceedings PDF, PMLR volume 306, retrieved 2026-10-07 · 31 pages