csprng

Introduction to CSPRNG (Cryptographically Secure Pseudo-Random Number Generator)

Overview

CSPRNG, short for "Cryptographically Secure Pseudo-Random Number Generator," is an extension for PyTorch, specifically designed with C++ and CUDA technologies. This toolkit empowers developers with advanced encryption capabilities and secure random number generation, ensuring both utility in cryptographic applications and compatibility with machine learning frameworks like PyTorch.

Key Features

1. AES 128-bit Encryption:

   • CSPRNG supports Advanced Encryption Standard (AES) 128-bit encryption/decryption.
   • Two modes are available: Electronic Codebook (ECB) and Counter (CTR).
   • These modes enhance data security, making it suitable for various confidential applications.

2. Cryptographically Secure Random Number Generation:

   • Provides a secure method for generating pseudorandom numbers.
   • Ensures that the output is unpredictable and suitable for use in cryptography.
   • Supports both CPU and GPU (CUDA) environments for enhanced performance.

Design Philosophy

CSPRNG employs a unique approach by generating a random 128-bit key on the CPU and utilizing AES 128-bit in CTR mode. This operation is efficiently executed on both CPUs and GPUs. A core advantage is its capability to handle parallel processing, ensuring efficient and secure random number generation even in high-demand scenarios.

Advantages

• Flexibility: Users can opt for seeded generators, useful for testing, or unseeded ones for full cryptographic security.
• Compatibility: A single generator instance can be adeptly used across both CPU and CUDA devices.
• Performance: The framework is designed for parallel computation even for CPU random number generation, which stands out compared to the standard PyTorch generator.

Torchcsprng API for Tensor Encryption/Decryption

CSPRNG's API introduces an intuitive way to handle tensor encryption and decryption, regardless of the data type. This enables encryption processes that cater to a wide range of applications, ensuring data security while maintaining flexibility.

• Encryption: The encrypt function allows users to secure tensors using AES 128-bit in ECB or CTR mode.
• Decryption: Similarly, the decrypt function ensures that encrypted data is safely and accurately restored to its original form.

Installation and Support

CSPRNG is versatile, supporting multiple operating systems such as Linux, macOS, and Windows. It's compatible with several versions of Python and PyTorch, making it accessible to a broad range of users and development environments.

Installation Methods

• Anaconda: Offers straightforward installation commands tailored for specific CUDA versions or CPU environments.
• Pip: Provides another avenue for installation, catering to various environments and prerequisites.

Getting Started

For developers eager to dive into CSPRNG, starting with the API is simple:

• Import the module alongside PyTorch:

  import torch
  import torchcsprng as csprng
  

• Create secure pseudo-random generators for applications needing immediate random device-generated security:

  urandom_gen = csprng.create_random_device_generator('/dev/urandom')
  

• Utilize the generator in CUDA environments for tasks such as initializing tensors with randomized values.

Community and Contribution

CSPRNG is an open-source project driven by community contribution. Developers are encouraged to contribute bug fixes and propose new features through discussions with the maintainers. The project operates under a BSD 3-clause license, promoting both collaboration and innovation.

Conclusion

CSPRNG offers a robust, secure solution for random number generation and encryption within the PyTorch landscape. Whether for cryptographic applications or enhancing machine learning workflows, it provides a seamless, efficient experience for developer needs.