SuperMask — UC3M Final Degree Project

UC3M · 4.º curso · Trabajo de fin de grado

Investigación sobre algoritmos evolutivos para buscar supermáscaras en redes neuronales, con experimentos sobre MNIST, CIFAR-10 y Animals-10.

Tecnologías: Python, PyTorch, algoritmos evolutivos.

Ver todos mis proyectos académicos

Author: Marcos Caballero Cortés

Degree: BSc in Computer Science

University: Universidad Carlos III de Madrid (UC3M)

---

Overview

SuperMask is a research project exploring evolutionary algorithms in deep neural networks to search for Supermasks.

This repository includes experimental code and results developed as part of a Final Degree Project (TFG) at UC3M.

---

Repository Structure

SuperMask/
├───1. Initial_resolutions/ 
|
├───2. Proving_LTH/         # Scripts for proving LTH
│
├───3. First_AE/            # First evolutionary algorithm
│
├───4. Second_AE/           # Second evolutionary algorithm
│
├───data/                   # Datasets
│
├───Scripts/                # Configuration files and first experiments
|
├── requirements.txt        # Python dependencies
|
├── LICENSE                 # MIT license        
|
└── README.md               # This file

---

Installation

We recommend using a virtual environment.

git clone https://github.com/cabamarcos/SuperMask.git
cd SuperMask

python3 -m venv venv
source venv/bin/activate    # or .\venv\Scripts\activate on Windows

pip install --upgrade pip
pip install -r requirements.txt

---

How to Run Experiments

All training and evaluation scripts are inside the folder of each experimental phase.

You can run the experiments via Jupyter notebooks:

jupyter notebook path/to/notebook.ipynb

The notebooks are self-contained and can be executed step-by-step from within the Jupyter interface.

---

License

This project is licensed under the MIT License.

---

Contact

For questions, suggestions, or collaborations:

• ✉️ [marcoscaballero.contacto@gmail.com]
• 🐛 Open an Issue