Unambiguous Representations in Neural Networks: A Relational Structure Approach to Consciousness

This directory contains the LaTeX source for a journal paper suitable for submission to Entropy or similar journals.

Structure

The paper is divided into modular sections for easier editing and collaboration:

• main.tex - Main document using MDPI template (for Entropy journal)
• main-article.tex - Alternative version using standard article class (easier to compile)
• references.bib - BibTeX bibliography
• sections/ - Individual section files:
  • 00-abstract.tex - Abstract placeholder (abstract is in main.tex preamble)
  • 01-introduction.tex - Introduction
  • 02-theory.tex - Theoretical framework
  • 03-methods.tex - Methods
  • 04-experiment1.tex - Experiment 1: MNIST digit classification
  • 05-experiment2.tex - Experiment 2: Spatial position decoding
  • 06-ambiguity-analysis.tex - Quantifying representational ambiguity
  • 07-discussion.tex - Discussion and related work
  • 08-conclusion.tex - Conclusion

Compilation

Option 1: Standard Article Class (Recommended for initial editing)

cd paper
pdflatex main-article.tex
bibtex main-article
pdflatex main-article.tex
pdflatex main-article.tex

Option 2: MDPI Template (For final submission to Entropy)

1. Download the MDPI LaTeX template from: https://www.mdpi.com/authors/latex
2. Extract the mdpi.cls file to a Definitions/ directory in this folder
3. Compile:

cd paper
pdflatex main.tex
bibtex main
pdflatex main.tex
pdflatex main.tex

Using LaTeXmk (automatic compilation)

latexmk -pdf main-article.tex
# or
latexmk -pdf main.tex

Figures

The paper references figures in ../figures/ directory. Ensure all figure files are present:

• pixel-encoding-relational-structure.png
• decoder-validation-accuracy-training-paradigms.png
• mnist-model-validation-accuracies.png
• target-similarity-only-output-neurons.png
• ablation-study-neuron-count-performance.png
• perm_distances_no_dropout.png
• perm_distances_dropout.png
• architecture-transfer-evaluation.png
• cross_architecture_heatmap_accuracy.png
• gram_neuron_ablation_plot.png
• dataset-classification-accuracy.png
• umap-input-neuron-similarity.png
• input-neuron-distance-prediction-accuracy.png
• target-similarity-only-input-pixels.png
• varying-subset-size-input-pixels.png
• knn-kernel-similarity-vs-decoder-accuracy.png

Modular Structure Benefits

The paper is divided into separate files to:

1. Facilitate collaboration - Multiple authors can work on different sections simultaneously
2. Version control - Git diffs are cleaner when sections are separate files
3. Easier editing - Work on one section at a time without navigating a large file
4. Reusability - Individual sections can be reused or reorganized easily

Customization

To customize for different journals:

1. Edit the preamble in main.tex or main-article.tex
2. Adjust formatting, citation style, figure sizes as needed
3. The modular structure means section content doesn’t need to change

Abstract and Keywords

The abstract is defined in the preamble of main.tex (for MDPI template) and at the beginning of main-article.tex (for standard article class). Update there to modify the abstract.

Keywords are also defined in the same locations.

Citation Management

All references are in references.bib. Add new references there in BibTeX format. Citations in the text use \cite{key} format, where key matches the BibTeX entry key.

Contributing

When making changes:

1. Keep section files focused on their specific content
2. Maintain consistent formatting across sections
3. Test compilation after major changes
4. Update this README if structure changes

License

[Add appropriate license information]

Contact

[Add contact information]