Invited Speakers:
Approximate Homomorphic Encryption and Privacy Preserving Machine Learning
Jung Hee Cheon (Seoul National University / CryptoLab Inc.)
We introduce the approximate homomorphic encryption HEaaN, which supports fixed-point operations via addition, multiplication and rescaling of plaintext vectors. This feature significantly accelerates approximate arithmetic of real numbers (compared with bitwise or finite field arithmetics) and realizes machine learning on encrypted data. We then present the recent developments on fast implementation of HEaaN and its application to logistic regression, neural network, and statistics on encrypted data. We conclude with several real-world applications, ongoing standardization activities and some possible future directions.
End-to-End Learning with Fully Homomorphic Encryption in Memory
Tajana Šimunić Rosing (University of California, San Diego)
The increasing amount of data and the growing complexity of problems has resulted in an ever-growing reliance on cloud computing. However, many applications, most notably in healthcare, finance or defense, demand security and privacy which today's solutions cannot fully address. Fully homomorphic encryption (FHE) elevates the bar of today's solutions by adding confidentiality of data during processing. It allows computation on fully encrypted data without the need for decryption, thus fully preserving privacy. To enable processing encrypted data at usable levels of classic security, e.g., 128-bit, the encryption procedure introduces noticeable data size expansion — the ciphertext is much bigger than the native aggregate of native data types. In this talk, we present MemFHE which is the first accelerator of both client and server for the latest Ring-GSW (Gentry, Sahai, and Waters) based homomorphic encryption schemes using Processing In Memory (PIM). PIM alleviates the data movement issues with large FHE encrypted data, while providing in-situ execution and extensive parallelism needed for FHE’s polynomial operations. While the client-PIM can homomorphically encrypt and decrypt data, the server-PIM can process homomorphically encrypted data without decryption. Our server-PIM is pipelined and is designed to provide flexible bootstrapping, allowing two encryption techniques and various FHE security-levels based on the application requirements. We evaluate our design at various security-levels and compare it with state-of-the-art CPU implementations for Ring-GSW based FHE. Our system is up to 20k× (265×) faster than CPU (GPU) for FHE arithmetic operations and provides on average 2007× higher throughput than the state of the art while implementing learning algorithms with FHE.
Homomorphic Computing, with a Focus on Hardware-Accelerated Homomorphic Encryption
Rosario Cammarota (Intel)
Enabling processing encrypted data elevates the bar of confidentiality in existing security solutions and opens the front to new applications. It preserves individuals' privacy and market competitiveness while sustaining societal and economic growth through data sharing, collaboration, and artificial intelligence. Homomorphic Encryption (HE) is a unique family of cryptographic methods to process encrypted data. HE applications can reduce the risk of third-party data leakage at the processing node while preserving both data ownership and lifecycle. However, the performance gap that even the most efficient HE schemes hinder enabling meaningful HE applications and adopting the technology. HE applications can be a million times slower than the corresponding unencrypted applications on existing hardware architectures. Other barriers to adoption include the lack of development tools to reduce non-recurring engineering costs to build HE applications and the lack of international standards. Innovation in hardware architecture is the first step in bridging the performance gap and setting directions to enable meaningful technology adoption. In this talk, I will share Intel's innovation from theory to algorithms down to hardware architecture to allow the adoption of processing encrypted data with HE. Jointly with Microsoft, our Standards & Industry Organization and academic partners, we develop novel HE platforms comprehensive of revolutionary hardware, software libraries, development tools, and applications to make HE technologies accessible, performant, and cost-effective. We build an ecosystem that can sustain the exponential growth of HE-based technologies.
CircLayer — Circuit Optimization & Scheduling Made Easy
Hayim Shaul (IBM Research)
CircLayer is a circuit abstraction layer part of the HELayers end-to-end framework to write high level fully homomorphic encryption code. CircLayer lets the researcher apply optimizations directly on the circuit level. In addition it gives control on scheduling decisions made when executing the circuit. This makes CircLayer an ideal choice for researches and developers who develop new optimization and scheduling algorithms.
Codesigning the Next Generation of Intelligent Computing Systems
Kevin Barker (Pacific Northwest National Laboratory)
Convergence is driving the emergence of tightly integrated scientific workflows that combine physical simulation, machine learning, and analytics. However, such workflows are not well-served by existing computing capabilities that separate HPC and AI/ML computing paradigms into distinct ecosystems. While co-processor accelerators such as GPUs have been successfully applied to both HPC and AI/ML workloads, software stacks, programming languages, and programming models are still largely incompatible. Additionally, many scientific machine learning workloads exhibit characteristics that hamper their performance on GPU throughput-oriented architectures. In this talk, we discuss our vision of the codesign process, as well as ongoing efforts at Pacific Northwest National Laboratory in codesigning hardware and software stacks to support this notion of convergence and the next generation of scientific computing.