Neural Network Initialization Using Color-Aware Perlin Noise

Project Overview

This repository contains the implementation and improvement of neural network initialization using color-aware Perlin noise. The project consists of two main components:

1. Original Implementation (perlin_original_code.py)

• Custom implementation of the methodology described in the original paper
• Note: Results may differ from the original paper due to implementation details/code not shared in the publication
• Independent implementation focused on capturing the core concepts

2. Improved Version (perlin_improved_code.py)

• Enhanced implementation featuring the new ColorPerlinNoiseDataset class
• Significant improvements in performance and functionality
• Promising results that warrant further research and validation

Documentation

For a detailed technical report and analysis of the methodology, experiments, and results, please refer to: Neural Network Initialization Using Color-Aware Perlin Noise.pdf

Configuration

The grid size can be customized by modifying the N and M parameters in the code:

N = your_value  # Width of the grid
M = your_value  # Height of the grid

Training

The optimizer parameters were determined using Optuna for optimal performance. These values are pre-configured in the code for the provided dataset.

Future Development

The improved version shows potential for publication pending further validation and enhancement of the findings.