Writing study: From movement to cognitive maps: recurrent neural networks reveal how locomotor development shapes hippocampal spatial coding

ICLR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICLR 2026 oral writing corpus

Marco P Abrate, Laurenz Muessig, Joshua Bassett, Hui Tan, Francesca Cacucci, Thomas Wills, Caswell Barry · Official ICLR 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: Empirical analysis, diagnosis & auditing (primary) · Architecture & representation design.

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

23 physical pages

Before references / appendix

≈ 10.1 page-equivalents

Main figures / tables

5 / 0

Attached appendix figures / tables

9 / 4

Attached appendix span

≈ 9.9 page-equivalents

Before references / appendix: ≈ 43.91%References / intervening material: ≈ 13.04%Attached appendix: ≈ 43.04%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	≈ 1.68 pp.	7.3%
2 COMPUTATIONAL ANALYSIS OF POSTNATAL LOCOMOTOR DEVELOPMENT	3	≈ 1.19 pp.	5.17%
3 RESULTS	4	≈ 5.23 pp.	22.74%
4 DISCUSSION	9	≈ 1.27 pp.	5.52%
REFERENCES	11	≈ 3 pp.	13.04%
APPENDIX	14	≈ 0.03 pp.	0.13%
A LLM USAGE STATEMENT	14	≈ 0.11 pp.	0.48%
B EXPERIMENTAL DATASETS	14	≈ 0.2 pp.	0.87%
C LOCOMOTOR DEVELOPMENT CLUSTERING ANALYSIS	14	≈ 3.07 pp.	13.35%
D RECURRENT NEURAL NETWORK	17	≈ 0.98 pp.	4.26%
E SPATIAL TUNING ANALYSIS	18	≈ 1.61 pp.	7%
F COMPARISON WITH ALTERNATIVE MODEL	20	≈ 3.9 pp.	16.96%

All detected figure and table locations

Main text

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

Attached appendix

• Figure A1 · p. 15
• Table A1 · p. 16
• Table A2 · p. 16
• Table A3 · p. 17
• Figure A2 · p. 17
• Table A4 · p. 19
• Figure A3 · p. 21
• Figure A4 · p. 21
• Figure A5 · p. 22
• Figure A6 · p. 22
• Figure A7 · p. 23
• Figure A8 · p. 23
• Figure A9 · p. 23

Argument and organization

1. Behavioral stages to predictive RNN

   Derives a developmental curriculum from rat trajectories before testing emergent representations.

2. Model prediction, biological comparison, alternative explanations

   Links mixed spatial-direction tuning to recorded cells and tests controls for temporal sampling and training exposure.

Problem, claim and contribution

The introduction moves from known cell-development timelines to a specific missing mechanism. It explicitly positions the RNN as a functional approximation rather than a literal anatomical circuit.

Sections 1 and 3, pp. 1–4 · Main paper

Argument and section order

Empirical locomotion clustering precedes simulation and model training. This ordering explains where the curriculum comes from before presenting neural patterns as an outcome.

Section 2 and Figure 1, pp. 3–4 · Main paper

Figures, captions and visual explanation

Figure 2 combines architecture, convergence, predicted imagery, and emergent tuning. The grid-cell input is visibly conditional, preventing the enhanced adult variant from being mistaken for the shared baseline.

Figure 2 and Eq. 2, p. 5 · Main paper

Tables, comparisons and evidence

Consistent colors encode positional, directional, and conjunctive tuning; line style distinguishes model from recordings, and diamonds identify the adult variant. The caption specifies normalization, uncertainty, and corrected tests.

Figure 4, p. 8; Table A4, p. 19 · Main paper

The alternative-explanation figure compares decoding, lesion effects, representation geometry, and frame differences together. Matching sensory change while failing to match representation quality is more informative than a sole end-point score.

Section 3.3 and Figure 5, pp. 8–9 · Main paper

Appendix and reproducibility

Simulation parameter search and selected values are separated from RNN hyperparameters; supplementary controls retain the main figures' encoding. Exact metric definitions allow readers to understand how biological and artificial units were compared.

Tables A1–A3, pp. 16–17; Appendices E–F, pp. 18–23 · Main paper

What to reuse—and what not to infer

Transferable lesson: Build a mechanism narrative from measured experience to controlled simulation, then test competing explanations using several complementary diagnostics.

Boundary: Computational interventions do not establish biological causality. The adult grid-input model is trained from scratch with an external spatial signal; no hippocampal crawl-stage recordings are available. Selected neuron matches maximize correlation, and min–max normalization affects distributional comparisons.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-09-19

Reading status

Full text + selected visuals inspected

Main text

Pages 1–10, Sections 1–4

Appendix / supplement

Pages 14–23, appendices A–F including simulation, training, spatial metrics, likelihood analysis, and all supplementary figures

Visually inspected PDF pages

5, 8, 9

• ICLR 2026 final proceedings · 23 pages