NCI’s Dr Luis Bernardo Pulido-Gaytan says his research into privacy-preserving machine learning models will lead to greater trust in AI tools.
Homomorphic encryption (HE) allows encrypted data to be analysed without the need to decrypt it. These encrypted data are known as ciphertexts.
For Dr Luis Bernardo Pulido-Gaytan, HE is a valuable tool for data security, particularly in the quantum computing era.
Pulido-Gaytan is a postdoctoral research associate in the Cloud Competency Centre at the National College of Ireland, where he designs privacy-preserving machine learning cognitive models in cloud environments using HE cryptosystems.
“Privacy-preserving modelling represents a rapidly growing research area with many potential applications and substantial benefits,” Pulido-Gaytan tells SiliconRepublic.com.
He received his PhD in computer science from the CICESE Research Centre in Mexico and has been involved in several international research groups, undertaking research internships at various universities, including the University of Göttingen in Germany and the University of the Republic in Uruguay.
Here, he tells us more about HE and his data-privacy research.
Tell us about your current research.
Privacy issues in cloud computing environments are becoming increasingly important. Traditional security practices successfully protect stored and transmitted data by encryption, but not during data processing when decryption is required to retrieve the data, creating risks for sectors with stringent privacy regulations like healthcare, pharmaceutical, government, finance, genomics etc.
Homomorphic encryption (HE) can solve these problems by providing the client with computation on encrypted data. In this case, users can develop and execute applications involving sensitive datasets on untrusted shared infrastructures without disclosing underlying information.
Image: Luis Bernardo Pulido-Gaytan
Building on the principle of protecting computation and not just data, my research is focused on advancing privacy-preserving machine learning by comprehensively investigating the design of privacy-preserving neural network models for classifying encrypted information using lattice-based HE cryptosystems that enable secure data processing in cloud environments.
The homomorphic processing of cognitive models requires operations not supported by HE, so our efforts have focused on designing cryptographically compatible replacement non-linear functions to operate over ciphertexts.
This research emerged from our broader interest in harmonising the rapid expansion of data-driven technologies with the imperative to protect sensitive information.
In your opinion, why is your research important?
My research on privacy-preserving machine learning using lattice-based cryptography addresses a critical need in our increasingly data-driven world: balancing the pursuit of innovative and accurate analysis with the protection of confidential information.
By leveraging HE in neural network models, we enable computations to be performed directly on ciphertexts, thereby eliminating the need to expose underlying data. This approach is especially relevant in sectors governed by stringent privacy regulations, which are seeking to raise awareness of data security risks and privacy preservation.
Privacy-preserving machine learning can foster greater trust in AI-driven solutions, eg hospitals may safely implement advanced diagnostic tools on encrypted patient records, financial institutions can detect fraud on encrypted transaction histories, and smart cities can optimise transportation planning without compromising the location data of citizens.
In the longer term, I foresee this technology encouraging a new generation of secure, cloud-based AI services where robust data analysis and individual privacy can effectively coexist.
What inspired you to become a researcher?
I can’t remember the exact moment I decided to become a researcher; it feels like one curiosity-driven decision led naturally to another.
In retrospect, I realise I’ve always been a curious person. Since a young age, I was captivated by physics, often watching documentaries that delved into complex topics. I was fascinated at how certain experts could explain intricate ideas with such clarity – even if I didn’t fully understand them. It struck me that their depth of understanding came from years of dedicated investigation. I remember thinking how amazing it was that people could be so focused and passionate about such a specific corner of knowledge.
It was also the collective support around me – my family’s unwavering encouragement, the inspiring teachers I continue to admire, friends who motivated me at every turn, and supervisors who offered invaluable guidance.
This synergy of curiosity, mentorship and a robust support system created an ideal environment for me to pursue a career in research. It is this journey of exploring new questions and contributing to the expansion of knowledge that I find deeply fulfilling.
What are some of the biggest challenges or misconceptions you face as a researcher in your field?
Homomorphic encryption is a promising tool for data security against the quantum computer threat. Those lattice-based schemes based their security on the hardness of the ‘ring learning with errors’ problem, which is currently secure against attacks from quantum computers.
After years of being considered purely theoretical, HE has now become technically feasible for certain real-world applications. However, the encrypted evaluation of computationally intensive tasks – such as training complex machine learning models – remains an open challenge due to the significant increase in computational complexity compared to their unencrypted analogues.
Therefore, a major challenge lies in bridging the gap between cutting-edge cryptographic research and complex real-world applications.
A related misconception is that breakthroughs occur overnight; in reality, genuine passion and dedication, supported by solid funding and collaboration, are crucial for sustained progress and for translating theoretical advances into impactful, real-world solutions.
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.
