In a significant advancement for data security and machine learning, Vaultree has unveiled the VENum Stack (Vaultree Encrypted Numbers), an open-source framework that enables developers to perform machine learning (ML) on encrypted data without compromising performance or security. This release addresses longstanding challenges in processing sensitive information securely, marking a pivotal moment in the fields of cryptography and artificial intelligence.‍

The Challenge of Secure Data Processing

Traditionally, analysing sensitive data with AI / ML models necessitated decryption, exposing information to potential breaches and complicating compliance with data privacy regulations. While Fully Homomorphic Encryption (FHE) allows computations on encrypted data, existing FHE schemes often suffer from scalability and performance limitations, hindering their practical application in real-world scenarios.‍

Introducing the VENum Stack

Vaultree’s VENum Stack comprises two core components:

• VENumpy: An internal FHE library that facilitates secure and scalable ML operations on encrypted data.

• VENumML: A Python library built upon Vaultree’s proprietary encryption scheme, designed to integrate AI / ML capabilities with FHE seamlessly.‍

By combining these tools, the VENum Stack empowers developers, regardless of their cryptographic expertise, to execute advanced AI / ML tasks securely, ensuring data privacy without sacrificing performance.‍

Key Features of VENumML‍

VENumML offers a range of functionalities tailored for encrypted data processing:

• Linear Models: Implementations of linear and logistic regression, optimised using stochastic gradient descent (SGD).
• Time Series Analysis (Phineus): Tools for Fast Fourier Transform (FFT) and gradient descent on encrypted time series data.
• Deep Learning: Support for transformer architectures, with features for facial recognition applications.
• Graph Analysis: Planned inclusion of algorithms like PageRank for encrypted graph data analy sis.

‍These features enable the processing of various data formats — including images, tabular data, unstructured data, graphs, and time series — while maintaining encryption throughout the AI / ML pipeline.‍

Open-Source Commitment and Community Engagement‍

Vaultree has open-sourced VENumML to foster innovation, transparency, and collaboration within the developer community. By providing access to this technology, Vaultree encourages developers and researchers to contribute to the evolution of privacy-preserving AI / ML applications.‍

The VENumML repository is available on GitHub:

Vaultree / VENumML

Encrypted Machine Learning library that relies on the Next-Gen Vaultree Fully Homomorphic Encryption (NG-FHE) library, VENumpy.

VENumML is a Privacy Preserving Machine Learning (PPML) library designed for building and applying machine learning models on encrypted data. With Vaultree's VENumpy library, VENumML leverages fully homomorphic encryption (FHE) techniques to perform computations on encrypted data without decryption, ensuring data privacy throughout the machine learning workflow. This repo is available to install via PyPI, see the installation instructions below for further details.

Explore the VENumML Documentation to learn more about our tool. Visit our GitHub Repository to access the codebase or check out the demos showcasing the capabilities of VENumML.

VENumML Key Features

• Encrypted Machine Learning: Implement various machine learning models while keeping the underlying data encrypted.
• Homomorphic Encryption Support: Works with Vaultree's VENumpy library that provides FHE functionalities.
• Privacy-Preserving Predictions: Make predictions on encrypted data without revealing the original features.

Modules

The VENumML library is under active development and currently includes implementations for:

• Linear Models
  • Linear…

View on GitHub

Developers can explore the codebase, access documentation, and participate in ongoing discussions to enhance the framework’s capabilities.‍

Real-World Applications and Industry Impact‍

Vaultree’s VENum Stack is more than a technological advancement — it’s a practical solution to some of the most pressing challenges in data-intensive industries like financial services and healthcare. By enabling secure, encrypted machine learning, VENum empowers organisations to innovate responsibly while adhering to stringent privacy regulations. Below two examples of possible use cases:‍

Financial Services: Optimising Cash Management with Secure Forecasting‍

Managing sensitive data, like ATM transaction volumes, has traditionally required decryption, exposing organisations to risks of data breaches and regulatory penalties. With VENumML, Vaultree’s Phineus module introduces a groundbreaking solution: Privacy-preserving time series forecasting.

Using advanced techniques like Fourier Transforms and Linear Regression on encrypted data, Phineus enables financial institutions to:

• Predict ATM cash needs while keeping transaction data encrypted.
• Reduce operational costs and downtime by optimising cash replenishment schedules.
• Comply seamlessly with privacy regulations like GDPR and the Gramm-Leach-Bliley Act.‍

By securely analysing sensitive data, financial organisations can enhance decision-making without compromising on privacy.‍

Healthcare: Diagnosing Rare Diseases with Encrypted Transformers‍

Healthcare providers face a critical need to analyse sensitive patient data while maintaining strict compliance with regulations like HIPAA. Vaultree’s VENumML demonstrated its capability to address this with encrypted transformer models for natural language processing (NLP).

In a healthcare-focused demo, encrypted patient medical records were processed to identify rare diseases, such as Wilson’s disease. Key innovations included:

• Data Encryption via VENumpy, ensuring patient information remains secure throughout preprocessing, tokenization, and inference.
• Transformer Architectures are fine-tuned to analyse encrypted inputs, enabling advanced NLP tasks.
• Encrypted Inference to securely generate diagnostic predictions without exposing sensitive data.‍

This approach empowers healthcare providers to securely leverage cutting-edge AI / ML tools, unlocking the potential for AI-driven medical innovation at scale.‍

Future Developments‍

Vaultree is committed to expanding the capabilities of the VENum Stack. Upcoming releases aim to include additional AI / ML models and support for a broader range of data types, further empowering developers to create secure, privacy-preserving applications across various domains.‍

Conclusion

Vaultree’s introduction of the VENum Stack represents a significant leap forward in the integration of machine learning and data encryption. By open-sourcing VENumML, Vaultree not only provides a powerful tool for secure data processing but also invites the global developer community to participate in shaping the future of privacy-preserving machine learning.