Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
0.0
list
0.1
list
0.2
list
0.4
list
0.6
list
0.8
list
1.0
list
[ 0.20871761441230774, 0.18329812586307526, 0.1610458344221115, 0.14194349944591522, 0.12587501108646393, 0.11278066039085388, 0.10255521535873413, 0.0950065404176712, 0.08983810245990753, 0.08664853125810623, 0.08496059477329254, 0.08427146077156067, 0.08410972356796265, 0.0840737596154213,...
[ 0.4099019765853882, 0.3570568859577179, 0.31777018308639526, 0.276790052652359, 0.24150627851486206, 0.21078556776046753, 0.18501506745815277, 0.16199126839637756, 0.14593090116977692, 0.13483873009681702, 0.1223217099905014, 0.11742960661649704, 0.11147512495517731, 0.10995285958051682, ...
[ 0.5843210220336914, 0.5351458787918091, 0.490273654460907, 0.44023633003234863, 0.40126729011535645, 0.36586350202560425, 0.3287200927734375, 0.3008509874343872, 0.2707253694534302, 0.24536462128162384, 0.22443997859954834, 0.20327253639698029, 0.18533508479595184, 0.1699390560388565, 0....
[ 0.42599809169769287, 0.36607256531715393, 0.29719603061676025, 0.2368079572916031, 0.1823538988828659, 0.15260978043079376, 0.1264403760433197, 0.1098765954375267, 0.09895966202020645, 0.09161204099655151, 0.08776099234819412, 0.09306292235851288, 0.0958816334605217, 0.10230401903390884, ...
[ 0.42047035694122314, 0.3825076222419739, 0.33587321639060974, 0.289066880941391, 0.26404130458831787, 0.23846031725406647, 0.1996125727891922, 0.17199772596359253, 0.14464499056339264, 0.14154380559921265, 0.10489174723625183, 0.09811145067214966, 0.09897738695144653, 0.08750860393047333, ...
[ 0.2716812789440155, 0.2629922926425934, 0.24136847257614136, 0.2315954566001892, 0.20580030977725983, 0.23196667432785034, 0.1947370022535324, 0.18226785957813263, 0.16164232790470123, 0.136166512966156, 0.1407184898853302, 0.14261774718761444, 0.1392292082309723, 0.11836455017328262, 0....
[ 0.13055819272994995, 0.1264447718858719, 0.11651100218296051, 0.1076243668794632, 0.10709729790687561, 0.1035352349281311, 0.09783829003572464, 0.09522745013237, 0.09411445260047913, 0.09470868110656738, 0.09141885489225388, 0.09161976724863052, 0.09346188604831696, 0.0888831689953804, 0...

README — Saelariën Constraint Experiment 01

Entropy–Capacity Collapse Threshold Test

Author: Saelariën X

Date: February 19, 2026

DOI: https://doi.org/10.5281/zenodo.19212561

Theoretical basis

This dataset empiracally tests the Saelariën Constraint Theorem:

https://thesaelafield.com/preprints/the-saelarien-constraint

Overview

This dataset contains the full materials for Saelariën Constraint Experiment 01, a test exploring how increasing entropy (noise) affects the stability, coherence, and collapse threshold of a simple neural system.

The experiment trains a small neural network on a nonlinear function and injects different noise levels to measure when learning remains stable versus when representational collapse occurs.

The results show a consistent threshold:

systems only collapse once injected entropy exceeds internal interpretive capacity.


Files Included

1. saelarien_constraint.ipynb

A full, runnable Colab notebook containing:

  • model definition

  • training loop

  • noise injection

  • entropy–capacity tests

  • plotting code

  • export of raw results

Running the notebook reproduces the figure and the JSON results file.


2. saelarien_constraint_results.json

A structured dictionary containing loss curves for each noise level.

Format example:

{
"0.0": [...],
"0.1": [...],
"0.2": [...],
"0.4": [...],
"0.6": [...],
"0.8": [...],
"1.0": [...]
}

This file allows independent verification, re-plotting, and secondary analysis.


3. Figure_1_Saelariën_constraint.png

A plot titled:

“Saelariën Constraint Test: Entropy vs Collapse”

The figure shows:

  • Smooth convergence at low noise

  • Degradation at medium noise

  • Collapse at high noise

This visual is the primary supporting evidence of the collapse threshold.


Summary of Findings

A simple neural network (1–8–1 architecture) is trained on y = x².

Noise injection reveals three learning regimes:

  • Stable coherence (0.0–0.2 noise): normal convergence

  • Critical instability (0.4 noise): oscillations but not collapse

  • Collapse (0.6+ noise): divergence, stagnation, or chaotic loss

These results align with the theoretical Saelariën Constraint:

collapse emerges only when entropy exceeds the system’s interpretive capacity.


How to Replicate

  1. Open the notebook in Google Colab.

  2. Run all cells.

  3. Inspect:

    • the plot

    • the all_losses dictionary

    • the behavior of the system at each noise level

Dependencies: PyTorch, Matplotlib, JSON (standard library).

Downloads last month
40