Security, Trust & ComplianceConference50min
Beyond encryption at rest: computing on encrypted data
This talk explains Homomorphic Encryption, which enables computation on encrypted data without decryption. It covers core concepts, modern Fully Homomorphic Encryption, practical applications in confidential AI, healthcare, and finance, and current limitations such as performance, ciphertext size, and implementation complexity.
talk.summaryAiDisclaimer
Andrea Iacono
talkDetail.whenAndWhere
Wednesday, October 7, 17:50-18:40
TBA 8
talks.roomOccupancytalks.noOccupancyInfo
Cloud providers encrypt data at rest and in transit, but most applications still decrypt information before processing it. Homomorphic Encryption (HE) challenges this model by allowing computations to be performed directly on encrypted the server computes the result without ever seeing the inputs.
This talk introduces the fundamentals of HE, from early partially homomorphic cryptosystems to Craig Gentry's 2009 breakthrough, explaining core concepts like ciphertext noise, noise budgets, and bootstrapping in an accessible way. We'll cover how modern schemes like CKKS, BGV, and TFHE differ in what they compute efficiently, and explore practical use cases: confidential AI inference where a model produces predictions without seeing user inputs and financial services where credit scoring operates entirely on encrypted records.
We'll also discuss current limitations (performance overhead, ciphertext size, and developer complexity) and how hardware acceleration and compiler tooling are rapidly closing these gaps. Attendees will leave with a clear picture of where HE is practical today and where it is not, and concrete pointers to start experimenting with open-source libraries.
This talk introduces the fundamentals of HE, from early partially homomorphic cryptosystems to Craig Gentry's 2009 breakthrough, explaining core concepts like ciphertext noise, noise budgets, and bootstrapping in an accessible way. We'll cover how modern schemes like CKKS, BGV, and TFHE differ in what they compute efficiently, and explore practical use cases: confidential AI inference where a model produces predictions without seeing user inputs and financial services where credit scoring operates entirely on encrypted records.
We'll also discuss current limitations (performance overhead, ciphertext size, and developer complexity) and how hardware acceleration and compiler tooling are rapidly closing these gaps. Attendees will leave with a clear picture of where HE is practical today and where it is not, and concrete pointers to start experimenting with open-source libraries.