VENumML: Encrypted ML and Healthcare
Unlocking Innovation in Privacy-Sensitive Industries
Artificial Intelligence (AI) and, in particular, Machine Learning (ML) are revolutionising the healthcare industry, but a significant challenge persists: How can healthcare organisations leverage sensitive data without compromising privacy? After all, models are only as good as the data with which they are fed, and feeding them well requires adherence to compliance regulations in order to protect patient privacy.
Enter stage right, Vaultree’s latest innovation, VENumML (Vaultree Encrypted Numbers Machine Learning)! VENumML addresses this challenge head-on. Our latest Python library enables healthcare providers to build and deploy encrypted machine learning models, ensuring data privacy at every step of the process. Open-sourced and designed for real-world applications, VENumML is set to redefine secure innovation in privacy-sensitive industries like healthcare and many more.
A Real-World Use Case: Encrypted Transformers in Healthcare
To showcase VENumML’s capabilities, we’ve developed a practical demo highlighting how encrypted ML can address challenges in healthcare. Healthcare providers often need to analyse unstructured data, like medical charts, while maintaining strict confidentiality. Traditional ML approaches require data to be decrypted before processing, creating vulnerabilities when the data is sent for inference. Meanwhile, regulations like HIPAA and GDPR mandate that patient data remain secure at all times.
The Solution: VENumML allows ML models to work directly on encrypted data, without constantly needing to decrypt and re-encrypt. Using a transformer-based architecture, the demo showcases how medical records encrypted with homomorphic encryption can be processed to identify rare diseases like Wilson’s disease—a genetic disorder affecting the brain and liver.
Breaking Down the Demo
Let’s delve into the technical steps of the demo, which combines state-of-the-art natural language processing (NLP) with advanced encryption techniques:
- Encrypting the Data:
- Patient medical records, simulated for this demo, are encrypted using VENumPy, Vaultree’s homomorphic encryption library. This ensures that all data remains protected throughout the entire ML pipeline.
- The encryption process is seamless and scalable, capable of handling unstructured text and tabular data.
- Preprocessing and Tokenization:
- The encrypted data is tokenized, splitting the text into smaller units (tokens) that can be fed into the ML model.
- This process ensures compatibility between the encrypted data and the transformer architecture.
- Transformers on Encrypted Data:
- A transformer-based model, pre-trained on large corpora, is fine-tuned to analyse encrypted medical records. This model operates entirely on encrypted inputs, processing sensitive information without exposing it.
- The encrypted transformer captures complex language patterns, extracting key insights even when the data remains encrypted.
- Inference and Diagnosis:
- Using encrypted inference, the model identifies potential rare diseases based on the medical records. For instance, it can suggest Wilson’s disease as a possible diagnosis based on the encrypted patient symptoms and lab results.
- All computations, from input to output, are conducted securely, ensuring that no sensitive information is exposed.
Key Features Unlocked by VENumML
The demo highlights several transformative features of VENumML:
- End-to-End Encryption: Data remains encrypted during preprocessing, training, and inference, eliminating potential vulnerabilities and regulatory compliance issues.
- Compatibility with Complex Models: VENumML supports transformer architectures, enabling advanced (Natural Language Processing) NLP capabilities on encrypted text.
- Real-World Scalability: Unlike traditional homomorphic encryption solutions, VENumML minimises performance overhead, making it suitable for large-scale applications.
- Versatility: While the demo focuses on healthcare, the library can be applied to other industries, such as finance and government, where data privacy is also critical.
Get Hands-On with VENumML
Curious to try VENumML? Meet us on Github and check it out! This demo notebook, as well as others, will walk you through the steps of building and deploying an encrypted transformer. It’s designed for developers and researchers who want to explore the cutting edge of privacy-preserving AI. If that sounds like you, we’d love to hear from you! If there is anything you need, let us know via Github issues or our Support portal.
Ready to Transform Your Data Security?
Discover how VEDS can revolutionise your data sharing and collaboration. Contact us today to learn more or request a demo.