# QNBP

**Repository Path**: mirrors_Orange-OpenSource/QNBP

## Basic Information

- **Project Name**: QNBP
- **Description**: Learning Linear Block Codes with Gradient Quantization
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-09-18
- **Last Updated**: 2026-02-13

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Learning Linear Block Codes with Gradient Quantization

This repository contains the open-source implementation of the research presented in: L-A. Dufrène, Q. Lampin, G. Larue, "Learning Linear Block Codes with Gradient Quantization," accepted to IEEE Transactions on Communications, paper number TCOM-TLS-24-1319.R2.

Overview: This codebase implements a novel machine learning approach for designing linear block codes (such as LDPC codes) using gradient descent optimization. The key innovation is a gradient quantization technique that maintains discrete (binary) values for trainable weights throughout the training process, enabling the direct learning of code structures while preserving their discrete nature.

Main simulation file is: learning_codes_on_a_budget_quantized.py