Experiments of Cohesive-Convergence Group

Evidence of Cohesive-Convergence Groups in Neural Network Optimization

This repository contains source files for a paper titled "Evidence, Definitions and Algorithms regarding the Existence of Cohesive-Convergence Groups in Neural Network Optimization". The paper explores novel insights into the convergence dynamics of neural networks through the lens of cohesive-convergence groups.

Paper Overview

The paper addresses fundamental questions related to the convergence process of neural networks, particularly focusing on the emergence of cohesive-convergence groups during the optimization process. It presents concepts, definitions, and algorithms aimed at understanding the interplay between dataset structure and optimization outcomes.

Content

• load_cifar_script.py: The Python script of data loading.
• train-model.ipynb: The Jupyter notebook containing preparation steps for experiments.
• sampling.ipynb: The Jupyter notebook containing experiments.
• tmp/: Directory containing the result of experiments.
• paper.pdf: PDF file of the paper content.

Reproduction

• Step 1: Run train-model.ipynb for dataset spliting and model preparation.
• Step 2: Run sampling.ipynb for reproducting two experiments in the paper.