Homomorphic Encrytion – End-to-end encryption to sensitive data

Cybersecurity has become a critical concern for individuals, organizations, and governments worldwide. As technology advances and more data is stored online, the risks of data leaks, hacker threats, and unauthorized access to sensitive information continue to grow. Data leaks occur when sensitive information is exposed, either intentionally or accidentally, to unauthorized parties. These leaks can result from hacking, insider threats, misconfigured databases, or accidental disclosures. High-profile data leaks in recent years have exposed millions of users’ personal information, including names, emails, passwords, and financial details. Such breaches not only harm individuals but also damage the reputation and financial stability of organizations.
Sensitive Data stored in large amounts of all users are regularly decrypted so data analysis, machine learning, recommendations, and various mathematical operations can be made. When any operation is done on sensitive data, the data is first decrypted, operations are performed, and then the data is again encrypted back. We take the example of bank balance as sensitive information in this case. The hackers can, therefore use various methods to hack and get access to decrypted sensitive information.
I now explain the work done at Hewlett Packard Labs, Palo Alto, which prevents the attacks by encrypting the data end to end by Homomorphic Encryption. Homomorphic encryption is a cryptographic technique that allows computations to be performed on encrypted data without needing to decrypt it first. This means that sensitive data can remain confidential while still being processed, making it highly valuable for privacy-preserving applications. With the use of partial Homomorphic Encryption simple operations such as addition and subtraction can be performed on encrypted data, therefore removing the need to decrypt data. Complex operation can also performed using fully homomorphic encryption. In untrusted environments, such methods implemented prevent attacks. The major disadvantage of this algorithm is computation. We at HP research labs, build library using parallelization methods such as OpenMP, pthreads, OpenCL to use multicores of CPU and GPU to accelerate the computation. We tested on Pailer Encryption
and achieved gains of 90x on mulitcore CPU and more than 3000x using commodity GPUs. We were able to test in real environments and achieved comparative performance to current existing solutions. We did present our findings as white paper and want to bring light to these techniques so that further research can be done in this field.
About the Author Akash Sahoo is Senior AI Researcher at Samsung Research America, working on AI, Robotics, Speech Models, NLP and current PhD student working on AI in IoT Research at Texas A&M University. He hold 2 Patents with Samsung, and published 3 Research Papers and has cofounded 4 startups. Worked with impactful research labs like HP Research Labs on Homomorphic Encryption, High Performance Computing.