On August 21st, 2024, Vaultree’s Co-Founder and CEO, Ryan Lasmaili, hosted an engaging and insightful webinar, delving into the future of secure data enablement through Encrypted Machine Learning (EML). The session features a panel of distinguished experts, including:

• Sarah Armstrong-Smith: Chief Security Advisor at Microsoft
• Dr Jacob Mendel: Head of Cryptography and Cybersecurity at State Street LLC
• Melissa Duke: Vaultree AI/ML Engineering Lead
• Cathal Smyth: Data Scientist Lead at Vaultree

This 60-minute webinar focused on EML's innovative capabilities and explored the transformative potential of truly secure and scalable AI/ML operations for organisations previously hindered by security and compliance concerns.

In this blog post, we will cover the key insights, practical applications, and future implications discussed during the webinar and highlight how Vaultree’s EML technology is set to redefine the landscape of data security.

‍

For the on-demand version of this webinar - Click Here!

‍

Key Trends Shaping the Future of Machine Learning and Data Security

‍

The Growing Importance of Complete Data Security 

‍

The webinar began with a comprehensive discussion on the critical need for advanced data security in today’s digital world. Sarah Armstrong-Smith emphasised that as cyber threats evolve, so must our approaches to data protection. “The attack surface itself is growing exponentially,” she noted, highlighting the increasing amount of data generated daily and the corresponding rise in potential vulnerabilities.

Dr. Jacob Mendel reinforced this by discussing the financial sector’s rapid adoption of digital assets and the growing need for encrypted data processing. He pointed out, “The end game should be key management—ensuring that data is encrypted not only in transit and at rest but also in use.” This sentiment underscores the importance of securing data throughout its entire lifecycle, particularly in an era where digital wallets and mobile payments are becoming ubiquitous. These concerns, combined with compliance requirements, hinder certain organisations from fully adopting AI/ML technologies as they already struggle to maintain complete data security with current technologies. 

‍

Understanding and Advancing Fully Homomorphic Encryption (FHE)

‍

The discussion then turned to the foundational technology behind Vaultree Encrypted Machine Learning (EML)—Fully Homomorphic Encryption (FHE). Cathal Smyth provided an accessible explanation of FHE, illustrating how Vaultree’s advancements in this technology have enabled computation on encrypted data at a scale and speed suitable for real-world applications for the very first time. 

Melissa Duke expanded on this by discussing how Vaultree’s FHE algorithms have overcome these traditional barriers, stating, “Our advancements in FHE have made it possible to implement encrypted machine learning on a scale that was previously unimaginable,” emphasising the groundbreaking nature of Vaultree’s technology and its potential to revolutionise data security across industries.

‍

Real-World Applications and Impact 

The webinar's main thrust began when the Vaultree experts Melissa and Cathal showcased several practical applications of Vaultree’s EML technology through live demos, illustrating its practical applications across various industries.

‍

Encrypted Time Series Forecasting with Phineus

Cathal Smyth kicked off the demos by introducing Phineus, the most recent addition to the VENumML library, capable of encrypted time series forecasting, amongst other forms of persistently encrypted data analysis. In this instance, Phineus is designed to predict trends and patterns in data without ever decrypting it, making it a game-changer for Industries that handle sensitive data, such as healthcare and finance.

‍

Cathal demonstrated how Phineus can forecast future values from encrypted datasets, maintaining data security throughout the process. The model efficiently handled complex data, such as financial transactions and patient health records, to predict trends like stock prices or patient admission rates. The demo highlighted Phineus’s ability to perform accurate predictions on encrypted data, ensuring that organisations can leverage powerful machine learning insights while keeping sensitive information completely secure.

‍

Cathal explained, "Phineus leverages our Fully Homomorphic Encryption (FHE) algorithms to analyse trends in encrypted data, making it ideal for scenarios where data privacy is paramount. This model is particularly valuable for sectors like healthcare and finance, where secure forecasting can significantly impact decision-making."

One of the most impactful sections of Cathal’s demonstration of Phineus was the direct comparison with Meta’s Prophet, which displayed Phineus's accuracy despite operating on persistently encrypted data.

‍

Secure Facial Recognition with Encrypted Images

‍

Melissa Duke followed with a live demo that showcased the use of Vaultree’s EML technology in facial recognition—a critical application for law enforcement, security agencies, and even healthcare providers.

In this demo, Melissa used encrypted images to perform facial recognition. She presented an example where two images of the same person, taken at different ages and in different poses, were encrypted and then analysed by the system to confirm a match. This process was completed without ever decrypting the images, preserving the individual's privacy throughout.

Melissa emphasised the significance of this application, stating, "This technology is a breakthrough for privacy-conscious industries. By enabling facial recognition on encrypted images, we allow organisations to implement secure identification systems without the risk of exposing sensitive biometric data. This is especially crucial for law enforcement agencies that need to protect both the privacy of individuals and the integrity of their operations."

‍

Time Series Analysis for Healthcare with Phineus

‍

In another demonstration, Cathal Smyth showcased the application of Phineus in the healthcare sector, particularly for time series analysis of patient health data. He demonstrated how the model could predict patient outcomes by analysing encrypted physiological signals such as ECG or EEG data. The ability to perform this analysis on encrypted data is critical in maintaining patient confidentiality while still benefiting from advanced predictive analytics.

Cathal explained, "In healthcare, patient data security is non-negotiable. Phineus enables healthcare providers to perform predictive analysis on sensitive data, such as predicting the likelihood of hospital readmission or monitoring chronic conditions, all while keeping the data fully encrypted."

‍

Collaborations and Community Engagement

‍

Vaultree’s commitment to fostering innovation and collaboration was a recurring theme throughout the webinar. Melissa Duke invited researchers and technology enthusiasts to engage with Vaultree’s open-source projects, highlighting the importance of community-driven advancements in encryption technologies.

Sarah Armstrong-Smith and Dr. Jacob Mendel also emphasised the critical role that cross-industry collaboration plays in enhancing data security, particularly in high-stakes sectors like healthcare, government, and finance.

‍

Q&A Recap 

‍

Question 1: What industries stand to benefit the most from Encrypted Machine Learning?

‍

Jacob Mendel, Head of Cryptography and Cybersecurity at State Street, emphasised that several industries are primed to benefit significantly from EML technology.“Financial services, healthcare, law enforcement, and government are at the forefront. They all deal with sensitive data that requires the highest levels of security. Encrypted ML provides a way to leverage advanced analytics while protecting that data.”

‍

Sarah Armstrong-Smith added, "Beyond these obvious sectors, any industry with high-value intellectual property—such as pharmaceuticals, energy, and legal services—could also see substantial gains. The ability to collaborate securely without risking data exposure is a game-changer."

‍

Question 2: How does Vaultree’s approach to Encrypted Machine Learning differ from other technologies?

‍

Melissa Duke, AI/ML Engineering Lead at Vaultree, explained the unique approach Vaultree has taken to overcome the existing limitations of machine learning on encrypted data. "Our solution uses Fully Homomorphic Encryption (FHE) to allow computations to be performed directly on encrypted data without needing decryption," she said. "This not only eliminates security risks but also ensures real-time data processing. Traditional methods require data decryption at some point, which poses a potential vulnerability. Our approach maintains data encryption throughout the entire process, from data ingestion to model inference."

‍

Cathal Smyth further added, “Many existing technologies struggle with the trade-offs between security and performance. We’ve focused heavily on optimising performance to ensure that Encrypted ML is not only secure but also practical for real-world, large-scale deployments. Our Phineus model, for example, demonstrates that we can handle complex data types and deliver actionable insights at speed.”

‍

Question 3: What are the next steps for Vaultree in expanding the use of Encrypted ML?

‍

Cathal Smyth responded by highlighting that while Vaultree has made significant strides, there are still areas for development. “At present, our primary focus has been on supervised learning applications, such as time-series forecasting and image recognition. As we look ahead, we are working on expanding into unsupervised learning and more complex algorithms. Our goal is to broaden the range of machine learning tasks that can be performed securely on encrypted data.”

‍

Melissa Duke added, "We’re also continuously refining our algorithms to improve the speed and efficiency of encrypted computations. The current limitations mainly revolve around computational overhead and scalability, but we are confident that our ongoing research and collaboration with academic institutions and industry partners will lead to further breakthroughs."

‍

Question 4: How does the implementation of EML impact organisational performance and cost?

‍

Jacob Mendel addressed the concerns around performance and cost: "Historically, encryption has been seen as a trade-off between security and performance. However, with advances in FHE and EML, this trade-off is becoming less significant. Organisations can maintain a high level of security without sacrificing operational efficiency. Moreover, the long-term benefits, such as reduced risk of data breaches, compliance with global data regulations, and the ability to unlock new analytical capabilities, far outweigh the initial costs."

‍

Sarah Armstrong-Smith also shared her perspective from a corporate standpoint: "The reality is that the cost of a data breach—in terms of both financial loss and reputational damage—can be astronomical. Investing in technologies like EML that mitigate these risks is a smart business decision. Additionally, as the technology matures, we expect the costs associated with its implementation to decrease."

‍

Question 5: How can interested parties learn more about Vaultree and get involved?

‍

Ryan Lasmaili encouraged participants to engage with Vaultree Directly: “We invite anyone interested in our technology to visit our website or LinkedIn page, where you can find more resources, request a demo, or even collaborate on research. We’re open to partnerships and are always looking to work with forward-thinking organisations and individuals to drive this innovation forward."

‍

Additional Information and Future Webinars

‍

For those who missed the live session, the full conversation is available on-demand here. Stay tuned for our next webinar, where we will dive deeper into Vaultree's data-sharing solutions and explore more practical use cases.

‍