SparSamp Artifact Description

This repository contains the Artifact for the paper "SparSamp: Efficient Provably Secure Steganography Based on Sparse Sampling". SparSamp introduces a novel steganography scheme that leverages sparse sampling techniques to achieve efficient and provably secure information hiding.

Overview

The Artifact provides a Python implementation for encoding and decoding messages using the SparSamp method within a neural network model. The core functionalities are encapsulated in the `encode_spar` and `decode_spar` functions, which utilize probabilistic token generation based on a given context.

Key Features

• Provable Security: The scheme is backed by rigorous theoretical analysis, offering security proofs.
• Efficiency: SparSamp significantly reduces computational overhead.
• Practicality: Designed for real-world steganography applications.

Main Functions

1. Encoding: The `encode_spar` function encodes a message into the generated token.
2. Decoding: The `decode_spar` function decodes the message from the generated token.

Usage

The Artifact includes the necessary code to reproduce the experiments presented in the paper. Users can run `test_sparsamp()` in `main.py` as an example, ensuring a pretrained language model and message bits file are prepared beforehand.

Requirements

- Python 3.8
- PyTorch 2.2.2
- Transformers 4.41.2
- SciPy
- Additional dependencies may be required based on your environment.

Notes

- Ensure the model and context are properly set up before using the encoding and decoding functions.
- Adjust parameters such as `block_size` and `top_p` according to specific requirements.
- The code uses random sampling; results may vary across runs if not controlled with `random_seed`.

This description provides an overview of the Artifact's purpose, features, and usage instructions. If you need further details or modifications, feel free to ask!