stability-of-simple-neural-networks
This repository contains the code and analysis from my bachelor's thesis, where I explore the stability of simple neural network architectures by studying their weight optimization behavior and robustness under different training conditions.
https://github.com/pablolah/stability-of-simple-neural-networks
Science Score: 44.0%
This score indicates how likely this project is to be science-related based on various indicators:
-
✓CITATION.cff file
Found CITATION.cff file -
✓codemeta.json file
Found codemeta.json file -
✓.zenodo.json file
Found .zenodo.json file -
○DOI references
-
○Academic publication links
-
○Academic email domains
-
○Institutional organization owner
-
○JOSS paper metadata
-
○Scientific vocabulary similarity
Low similarity (9.8%) to scientific vocabulary
Repository
This repository contains the code and analysis from my bachelor's thesis, where I explore the stability of simple neural network architectures by studying their weight optimization behavior and robustness under different training conditions.
Basic Info
Statistics
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
- Releases: 0
Metadata Files
README.md
Stability of Simple Neural Network Architectures
This repository contains the code and analysis from my bachelor's thesis, Stability of Simple Neural Network Architectures. The project investigates the performance and stability of basic neural network architectures, analyzing how different design choices affect their robustness and learning dynamics.
Overview
Neural networks have become a fundamental tool in modern machine learning, but their stability—particularly in simple architectures—is not always well understood. In this study, I explore the behavior of small neural networks by: - Implementing and training simple feedforward networks in Python. - Examining how different architectures and activation functions impact learning. - Using mathematical constraints and contour plots to visualize network stability. - Analyzing how the learning rate influences the final weight configurations.
The goal is to gain insights into how small neural networks generalize and how their training dynamics lead to different optima.
Main Topics Covered
- Programming Neural Networks: Basic implementations in Python using TensorFlow and Keras.
- Interpolation with Simple Architectures: Training neural networks to approximate functions with noise.
- Stability Analysis: Using mathematical constraints and contour plots to assess how robust learned weight configurations are.
- Effect of Learning Rate: Investigating how varying learning rates influence the convergence behavior of neural networks.
Code Structure
- Jupyter notebooks with experimental results and visualizations.
- Includes the full thesis as a PDF for reference.
How to Use This Repository
This repository is meant as a reference for understanding stability in simple neural networks. The code is not guaranteed to be a plug-and-play framework but rather a collection of experiments that accompany my thesis. If you're interested in running the experiments, you may need to adapt the code to your setup.
Thesis Report
The full report, detailing the theoretical background and experimental results, can be found in the report/ directory.
Contact
If you have any questions or would like to discuss this work, feel free to reach out!
Owner
- Login: PabloLah
- Kind: user
- Repositories: 1
- Profile: https://github.com/PabloLah
Citation (CITATION.cff)
cff-version: 1.2.0 message: "If you use this software, please cite it as below." authors: - family-names: "Lahmann" given-names: "Pablo" orcid: "https://orcid.org/0000-0000-0000-0000" title: "Stability-of-Simple-Neural-Networks" version: 1.0.0 doi: 10.5281/zenodo.1234 date-released: 2021-12-28 url: "https://github.com/PabloLah/Stability-of-Simple-Neural-Networks"
GitHub Events
Total
- Push event: 11
Last Year
- Push event: 11