🎟️ Losing Tickets in Neural Representations

Exploring the Lottery Ticket Hypothesis in SIRENs

This repository contains the code and experiments for the Losing Tickets in Neural Representations project, which investigates the Lottery Ticket Hypothesis (LTH) within Sinusoidal Representation Networks (SIRENs).

📖 Read the Full Interactive Blog Post

Visual diagnostics, interactive PSNR curves, and detailed experimental analysis.

---

🚀 Installation & Setup

1. Clone the repository

git clone https://github.com/Neilus03/losingtickets.git
cd losingtickets

2. Install dependencies

pip install torch torchvision numpy pillow tqdm

3. Run the experiment

python main.py

This will:

• Train a baseline dense SIREN model on the Mona Lisa.
• Run the iterative pruning loop for Winner, Random, and Loser tickets.
• Generate reconstruction images and PSNR logs in the outputs/ directory.

📁 Project Structure

• src/: Core implementation (SIREN, pruning logic, training loop).
• docs/: GitHub Pages code and interactive blog.
• checkpoints/: Model state dicts for different pruning iterations.
• outputs/: Reconstructed images and benchmark results.

---

Author: Neil De La Fuente