HUTAO: A Reconfigurable Homomorphic Processing UniT With Cache-Aware Operation Scheduling

The secure and efficient processing of private information in the cloud computing paradigm is still an open issue. New security threats arise with the increasing volume of data into cloud storage, where cloud providers require high levels of trust, and data breaches are significant problems. Encrypting the data with conventional schemes is considered the best option to avoid security problems. However, a decryption process is necessary when the data must be processed, but it falls into the initial problem of data vulnerability. The user cannot operate on the data directly and must download it to perform the computations locally. In this context, Fully Homomorphic Encryption (FHE) is considered the holy grail of cryptography in order to solve cybersecurity problems, it allows a non-trustworthy third-party resource to blindly process encrypted information without disclosing confidential data. FHE is a valuable capability in a world of distributed computation and heterogeneous networking. In this survey, we present a comprehensive review of theoretical concepts, state-of-the-art, limitations, potential applications, and development tools in the domain of FHE. Moreover, we show the intersection of FHE and machine learning from a theoretical and a practical point of view and identify potential research directions to enrich Machine Learning as a Service, a new paradigm of cloud computing. Specifically, this paper aims to be a guide to researchers and practitioners interested in learning, applying, and extending knowledge in FHE over machine learning.KeywordsCloud securityFully homomorphic encryptionMachine learning as a service

Privacy-preserving machine learning (PPML) demands secure inference on sensitive data in untrusted environments. Fully homomorphic encryption (FHE) enables computation directly on encrypted data, but it incurs prohibitive communication and computational overhead on edge devices. Hybrid homomorphic encryption (HHE) mitigates this by combining symmetric encryption with FHE, reducing client-side costs; however, existing implementations remain impractical for resource-constrained deployments. This paper proposes the first end-to-end hardware-accelerated HHE framework, which integrates a lightweight symmetric cipher optimized for FHE compatibility with a dedicated hardware accelerator. Beyond this integration, we introduce a microarchitectural optimization for throughput and energy efficiency. The proposed architecture is validated within a complete PPML pipeline, demonstrating significantly lower latency and power consumption than software implementations. Experimental results on a PYNQ-Z2 platform with MNIST validate our approach, achieving a 50× reduction in client-side encryption latency and a 2× throughput gain over existing FPGA-based HHE accelerators. These improvements enable practical and secure inference on edge devices. Our contributions establish a hardware-software co-design methodology for deploying scalable, secure machine learning in resource-constrained environments, validating the feasibility of low-power HHE for edge deployment.

Secure report is the problem of a client that retrieves all records matching specified attributes from a database table at the server (e.g. cloud), as in SQL SELECT queries, but where the query and the database are encrypted. Here, only the client has the secret key, but still the server is expected to compute and return the encrypted result. Secure report is theoretically possible with Fully Homomorphic Encryption (FHE). However, the current state-of-the-art solutions are realized by a polynomial of degree that is at least linear in the number m of records, which is too slow in practice even for very small databases. We present the first solution that is realized by a polynomial that attains degree independent of the number of records m, as well as the first implementation of an FHE solution to Secure report. This is by suggesting a novel paradigm that forges a link between cryptography and modern data summarization techniques known as coresets (core-sets), and sketches in particular. The key idea is to compute only a coreset of the desired report. Since the coreset is small, the client can quickly decode the desired report that the server computes after decrypting the coreset. We implemented our main reporting system in an open source library. This is the first implemented system that can answer such database queries when processing only FHE encrypted data and queries. As our analysis promises, the experimental results show that we can run Secure report queries on billions records in minutes on an Amazon EC2 server, compared to less than a hundred-thousands in previous FHE based solutions.

Data privacy is a growing concern as more cloud-based solutions for extracting insights from images are available. Fully Homomorphic Encryption (FHE) is one of the state-of-the-art techniques to enable privacy preserving machine learning. However, encrypted deep neural network-based inference methods face the fundamental challenge of optimizing computational depth and complexity to achieve an acceptable trade-off between accuracy and practical feasibility. Existing works only report high level implementations without providing rigorous analysis of the intricacies involved in the FHE implementation of generic primitive operators of a convolutional neural network (CNN). In this paper, we use the CKKS encryption scheme available in the open-source HElib library to run encrypted inference experiments on the MNIST dataset. The experiments indicate that efficient ciphertext packing schemes, model optimization and multi-threading strategies play a critical role in determining the throughput and latency of the inference process. We also show that operational parameters of the chosen FHE scheme such as the degree of the cyclotomic polynomial, depth limitations of the underlying leveled HE scheme, and the computational precision parameters result in significant trade-offs between accuracy, security level and computational time of the machine learning model. The key contribution of the paper is the analysis and recommendation of optimization techniques for efficient encrypted CNN inference.

Fully homomorphic encryption (FHE) is a post-quantum secure cryptographic technology that enables privacy-preserving computing on an untrusted platform without divulging any secret or sensitive information. The core of FHE is the bootstrapping algorithm, which is the intermediate refreshing procedure of a processed ciphertext. However, this step has been the computational bottleneck that prevents real-world deployments among various FHE schemes. This paper, to the best of our knowledge, for the first time, presents a scalable and ultra-highly parallel design for the number theoretic transform (NTT)-based polynomial multiplier with a variable number of reconfigurable processing elements (PEs). Hence, the highest degree of acceleration can be achieved for any targeted hardware platform by implementing as many PEs as possible under the resource constraint. The corresponding addressing and scheduling schemes are also proposed to avoid memory access conflict for the PEs, which yields an extremely high utilization ratio of 99.18% on average. In addition, the latency of the proposed design with the general negative wrapped convolution algorithm is reduced by 59.20% compared to prior works.

With the wide adoption of cloud computing paradigm, it is important to develop appropriate techniques to protect client data privacy in the cloud. Encryption is one of the major techniques that could be used to achieve this goal. However, data encryption at the rest alone is insufficient for secure cloud computation environments. There is also the need for efficient techniques to carry out computation over encrypted data. Fully homomorphic encryption (FHE) and garbled circuits are naturally used to process encrypted data without leaking any information about the data. However, existing FHE schemes are inefficient for processing large amount of data in cloud and garbled circuits are one time programs and cannot be reused. Based on quaternion/octonion algebra and Jordan algebra over finite rings \(\mathbb {Z}_q\), this paper designs efficient fully homomorphic symmetric key encryption (FHE) schemes without bootstrapping (that is, noise-free FHE schemes) that are secure in the weak ciphertext-only security model assuming the hardness of solving multivariate quadratic equation systems and solving univariate high degree polynomial equation systems in \(\mathbb {Z}_q\). The FHE scheme designed in this paper is sufficient for privacy preserving computation in cloud.

Fully Homomorphic Encryption (FHE) is a cryptographic method that guarantees the privacy and security of user data during computation. FHE algorithms can perform unlimited arithmetic computations directly on encrypted data without decrypting it. Thus, even when processed by untrusted systems, confidential data is never exposed. In this work, we develop new techniques for accelerated encrypted execution and demonstrate the significant performance advantages of our approach. Our current focus is the Fully Homomorphic Encryption over the Torus (CGGI) scheme, which is a current state-of-the-art method for evaluating arbitrary functions in the encrypted domain. CGGI represents a computation as a graph of homomorphic logic gates and each individual bit of the plaintext is transformed into a polynomial in the encrypted domain. Arithmetic on such data becomes very expensive: operations on bits become operations on entire polynomials. Therefore, evaluating even relatively simple nonlinear functions with the CGGI cryptosystem, such as a sigmoid, can take thousands of seconds on a single CPU thread. Using our novel framework for end-to-end accelerated encrypted execution called ArctyrEX, developers with no knowledge of complex FHE libraries can simply describe their computation as a C program that is evaluated 18x faster on average relative to the GPU-accelerated Concrete library for multiplication-intensive benchmarks.

With recent development in internet speed and reliability, cloud computing has become a more reliable solution for the user. In many cases where data privacy is critical, fully homomorphic encryption (FHE) can be a security solution for securing cloud computing. FHE enables computation on encrypted data, hence it ensures data privacy in case of cloud computing. One popular scheme of FHE is the BFV homomorphic encryption scheme, which is based on ring learning with error (RLWE) computation. The BFV scheme uses ring polynomials as the main object, hence its encryption, decryption, and evaluation require high-degree polynomial multiplication. In this paper, we present comprehensive design and implementation of a hardware architecture to accelerate encryption and decryption in BFV scheme. Our accelerator uses convolution approach for calculating a polynomial multiplication. To implement the convolution, we use a systolic array to calculate polynomial convolution followed by a simple delayed subtraction to calculate polynomial modulo reduction inside our accelerator’s core. Moreover, we use a built-in Gaussian pseudo-random number generator (PRNG) to generate Gaussian noise in the encryption operations. Finally, we implement the 1024 degrees BFV accelerator on the Xilinx PYNQ Z1 board and compare the encryption and decryption performances to other methods as well as a software implementation on Intel Core i7 with 8GB memory. Experimental results show that our accelerator outperforms the clock cycles of other methods with the same polynomial degrees 1024 up to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$22\times $ </tex-math></inline-formula> . Moreover, our proposed Gaussian PRNG has better <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2\times $ </tex-math></inline-formula> correlation compared to the rotation-only-based PRNG. Finally, our accelerator accelerates up to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$9\times $ </tex-math></inline-formula> for encryption and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$3.5\times $ </tex-math></inline-formula> for decryption as well as <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$6.8\times $ </tex-math></inline-formula> for overall compared to Microsoft SEAL on Intel Core i7 processor with 8GB memory. The proposed design is scalable for higher degrees polynomial multiplication and useful for security technology such as high-speed secure cloud computing, blind computing, and secure communication.

Healthcare data has risen as a top target for cyberattacks due to the rich amount of sensitive patient information. This negatively affects the potential of advanced analytics and collaborative research in healthcare. Homomorphic encryption (HE) has emerged as a promising technology for securing sensitive healthcare data while enabling computations on encrypted information. This paper conducts a background survey of HE and its various types. It discusses Partially Homomorphic Encryption (PHE), Somewhat Homomorphic Encryption (SHE), Fully Homomorphic Encryption (FHE) and Fully Leveled Homomorphic Encryption (FLHE). A critical analysis of these encryption paradigms’ theoretical foundations, implementation schemes, and practical applications in healthcare contexts is presented. The survey encompasses diverse healthcare domains. It demonstrates HE’s versatility in securing electronic health records (EHRs), enabling privacy-preserving genomic data analysis, protecting medical imaging, facilitating privacy-preserving machine learning (ML), supporting secure federated learning, ensuring confidentiality in clinical trials, and enhancing remote monitoring and telehealth services. A comprehensive examination of potential vulnerabilities in HE systems is conducted. The research systematically investigates various attack vectors, including side-channel attacks, key recovery attacks, chosen plaintext attacks (CPA), chosen ciphertext attacks (CCA), known plaintext attacks (KPA), fault injection attacks (FIA), and lattice attacks. A detailed analysis of potential defense mechanisms and mitigation strategies is provided for each identified threat. The analysis underscores the importance of HE for long-term security and sustainability in healthcare systems.

Cloud service capabilities can always be clearly indicated by Quality of Service (QoS) properties. The quality and effectiveness of cloud services can be measured at both client and server sides. The parameters such as response time, throughput and failure rate influence the QoS properties at the client side. These parameters may differ with the utilization of the application for the same service. The proposed Bayesian Personalized Ranking-based Rank Prediction Scheme (BPR-RPS) provides rank prediction of QoS properties on the client side, since client-side properties vary with the usage of the application. The proposed rank prediction framework provides ranking to the services when particular cloud services are requested by the client. Personalization of cloud providers and customers is made possible by concentrating on the item recommendation imposed on the cloud by creating user-specific ranking for a set of items. The past response of the user with respect to the item aids in decision-making. The proposed ranking framework includes the following modules: Computation of similarity index between the active and ad hoc users and finding of identical users based on similarity index values.

As the computation bottleneck in lattice-based cryptography (LBC), the polynomial multiplication based on number theoretic transform (NTT) has been continuously studied for flexible hardware implementations with high area-efficiency. This paper presents an area-efficient and configurable NTT-based polynomial multiplier (AC-PM) incorporating algorithmic and architectural level optimization techniques. For the core operation of polynomial multiplication, two low-complexity and fast modular multiplication algorithms are introduced with loose constraints of LBC-friendly primes. Based on the proposed algorithms, a reconfigurable processing element (RPE) is dedicatedly designed to execute all the operations in an NTT-based polynomial multiplication: NTT, inverse NTT (INTT), and coefficient-wise multiplication (CWM). The proposed AC-PM can be configured with different numbers of RPEs and supports various polynomial degrees without recompilation. Additionally, the dataflow complexity is greatly simplified. More importantly, to the best of our knowledge, the twiddle factors are reused, for the first time, to support both NTT and INTT with multiple polynomial degrees, which leads to increased flexibility of AC-PM with small overhead on hardware resource. FPGA implementation results demonstrate that the proposed AC-PM significantly outperforms the prior arts in both flexibility and area efficiency.

Over the past few years, machine learning has been responsible for the rapid advancements in fields such as computer vision, natural language processing and speech recognition. No small part of this success is due to data becoming more and more available, often being collected in privacy-invasive ways. The aim of this work is to study the use of two privacy-preserving solutions for training machine learning models: Federated Learning (FL) and Secure Multiparty Computation (MPC). Federated learning is a subfield of machine learning that allows training models on a large, decentralized corpus of data residing on edge devices like smartphones. Instead of sharing data, users collaboratively train a model by only sending weight updates to a server. By leveraging secure multiparty computation, it can be ensured that the server cannot inspect any individual user's update. To assess the feasibility of these approaches in different settings, a client-server architecture was implemented in Python and multiple experiments were run on datasets made available by LEAF in order to investigate ways of improving the overall performance of the models trained in a federated manner.

One of the challenges faced by machine learning in human activity recognition systems is the different distributions of the training and test samples. Transfer learning constitutes a solution to this problem. On the other hand, to perform transfer learning, it is necessary to have access to the original dataset. However, access to the dataset to implement the transfer learning algorithms results in a privacy breach. To deal with this challenge, this paper presents semi-supervised and unsupervised scenarios for privacy-preserving transfer learning in centralized and distributed manner. In the proposed distributed algorithms, it is not necessary to share the original data among the clients to implement the transfer learning algorithms. Instead, the transfer learning process can be fulfilled without having the original datasets. PPSETR and PPUSTR algorithms transfer the knowledge while preserving the privacy of the datasets on the client side. In contrast, PPDSETR and PPDUSTR algorithms provide the privacy protection of the distributed data on both the client and server sides. The proposed semi-supervised algorithms reduce the recognition error rate by 20.58% and the unsupervised algorithms decrease the recognition error rate by about 15.97% while these algorithms considerably preserve the privacy.

At IHEP (Institute of High Energy Physics, Chinese Academy of Sciences), computing resources are contributed by different experiments including BES, JUNO, DYW, HXMT, etc. The resources were divided into different partitions to satisfy the dedicated experiment data processing requirements. IHEP had a local Torqu&amp;Maui cluster with 50 queues serving for above 10 experiments. The separated resource partitions leaded to imbalance resource load. In a typical situation, BES resource partition was quite busy without free slot but still with lots of jobs in idle, while JUNO resources are free and wasted seriously. After moving resources from Torque&amp;Maui to HTCondor in 2016, job scheduling efficiency has been improved a lot. In order to balance the imbalance resource load, we designed an efficient sharing strategy to improve the overall resourceutilization. We created an unified pool shared by all experiments. For each experiment, resources are divided into two parts: dedicated resource and sharing resource. The slots in dedicated resource only run jobs from its own experiment, and the slots in sharing resource are shared by jobs from all experiments. Default ratio of dedicated resource to sharing resource is 1:4. To maximize the sharing effectiveness, the ratio is dynamically adjusted between 0:5 and 4:1 based on the number of jobs submitted by each experiment. We have developed a central control system to decide how many resources can be allocated to each experiment group. This system is implemented at two sides: server side and client side. A management database is built at server side, which is storing resource, group and experiment information. Once the sharing ratio needs to be adjusted, resource group will be changed and updated into database. The resource group information is published to the server buffer in real-time. The client periodically pulls resource group information from server buffer via https protocol And resource scheduling configuration at client side is changed based on the resource group information. With this method, share ratio can be modified and deployed dynamically. Sharing strategy is implemented with HTCondor. ClassAd mechanism and accounting-group in HTCondor facilitate to utilizethe sharing strategy at IHEP computing cluster. With the sharing strategy, resource usage has been improved dramatically.

In recent years, the type and quantity of news are growing rapidly, and it is not easy for users to find the news they are interested in the massive amount of news. A news recommendation system can score and predict the candidate news, and finally recommend the news with high scores to users. However, existing user models usually only consider users’ long-term interests and ignore users’ recent interests, which affects users’ usage experience. Therefore, this paper introduces gated recurrent unit (GRU) sequence network to capture users’ short-term interests and combines users’ short-term interests and long-term interests to characterize users. While existing models often only use the user’s browsing history and ignore the variability of different users’ interest in the same news, we introduce additional user’s ID information and apply the personalized attention mechanism for user representation. Thus, we achieve a more accurate user representation. We also consider the risk of compromising user privacy if the user model training is placed on the server side. To solve this problem, we design the training of the user model locally on the client side by introducing a federated learning framework to keep the user’s browsing history on the client side. We further employ secure multi-party computation to request news representations from the server side, which protects privacy to some extent. Extensive experiments on a real-world news dataset show that our proposed news recommendation model has a better improvement in several performance evaluation metrics. Compared with the current state-of-the-art federated news recommendation models, our model has increased by 0.54% in AUC, 1.97% in MRR, 2.59% in nDCG@5%, and 1.89% in nDCG@10. At the same time, because we use a federated learning framework, compared with other centralized news recommendation methods, we achieve privacy protection for users.