A lightweight, privacy-preserving, and verifiable SVM training scheme designed for resource-constrained clients, which maintains stable classification performance while achieving an order-of-magnitude decrease in training runtime compared with existing ciphertext-based methods.
Abstract
The outsourcing of machine learning tasks, such as Support Vector Machine (SVM) training, to cloud platforms poses significant security challenges, primarily concerning the confidentiality of sensitive training data and the integrity of computation results returned by potentially malicious servers. To address these challenges, this paper proposes a lightweight, privacy-preserving, and verifiable SVM training scheme designed for resource-constrained clients. Our scheme leverages a replicated secret sharing protocol to securely distribute training data and model parameters across multiple non-colluding servers, executing the entire collaborative training process in the share domain without leaking plaintext information. Furthermore, to guarantee computational correctness, we introduce a novel interval-based index point storage strategy combined with a bilinear mapping-based parameter label consistency check. This verifiable mechanism enables clients to perform sampled, lightweight audits of the cloud’s intermediate training states and final outputs. Experimental evaluations on multiple typical datasets demonstrate that the proposed scheme maintains stable classification performance while achieving an order-of-magnitude decrease in training runtime compared with existing ciphertext-based methods, offering a highly configurable trade-off among verification coverage, computational overhead, and storage cost.
Multi-Key Searchable Encryption (MKSE) enables data owners (DOs) to outsource their data to a cloud server (CS) while supporting fine-grained data sharing with other authorized users. Most existing MKSE schemes can protect data users’ (DUs’) search query privacy against collusion attacks between malicious DOs and the CS. However, the CS is not fully trusted and may maliciously return forged or incomplete search results. To address this issue, Verifiable MKSE (VMKSE) is proposed by leveraging Garbled Bloom Filter (GBF), which can support verifiability even when the search results are empty. Unfortunately, due to the massive native storage redundancy of GBF, the storage and computational overhead of verification evidence generated in the sharing phase increases as the number of shared documents grows. Therefore, in this paper, we present a novel VMKSE scheme (VMKSE-BFF) by adopting BFF, which can simultaneously support verifiability of and secure data sharing in a multi-user setting. We provide a comparison with the existing VMKSE schemes. Experimental results on a real-world dataset show a significant performance improvement of VMKSE-BFF.
Yandong Su, Bing-Hang Wang, Yan-Jie Xiang et al.· Mathematics· 0 citations
This work proposes PRoVeFL-a novel, modular FL framework that is Privacy-preserving, Byzantine-Robust, and ensures Verifiable aggregation, and improves runtime over the prior works, Prio and ELSA, based on distributed trust with comparable security guarantees, up to 100x and 10x, respectively.
Harsh Kasyap, Anil Kumar Pradhan, U. Atmaca et al.· 0 citations
The growing demand for large models in industrial Internet necessitates secure sharing of cloud-stored training data. However, existing Ciphertext-Policy Attribute-Based Encryption schemes suffer from low efficiency and insufficient accuracy in tracing key-leaking users. We propose a traceable secure data sharing scheme based on credibility, which enforces dual access control via credibility and attributes. It strengthens the binding of user and key and integrates a large model to create an intelligent traceability framework, utilizing multi-layer query mechanisms and Merkle trees to reduce redundant verification operations and improve tracing efficiency. Under the DBDH assumption, the scheme is IND-CPA secure. Experiments on the “Insider Threat Test Dataset” demonstrate a traceability accuracy of 99.46%, significantly reducing unnecessary verification operations. The proposed scheme provides an efficient, accurate, and practical solution for secure data sharing and malicious user traceability in distributed industrial cloud environments.
Shaojian Zhang, Jun Ye· 2026 5th International Confe...· 0 citations
Decision tree classification serves as a fundamental component in many machine learning applications. As inference services are increasingly outsourced to cloud platforms, designing privacy-preserving mechanisms has become important. To secure outsourced decision tree inference, Chen et al. proposed SecDT, an efficient secret-sharing-based framework. While SecDT offers notable efficiency and preliminary security guarantees, we identify a critical vulnerability: attribute leakage. Specifically, the attributes associated with decision tree nodes are exposed to non-owner parties, which risks revealing sensitive model information. In this paper, we first present two secure enhancements, SecDT+v1 and SecDT+v2. These variants use the transformation matrix technique to obfuscate node attributes within the secret-shared domain. To prevent more advanced leakage through attribute access patterns, we design two advanced variants, SecDT+vH and SecDT+vDP, which incorporate dot-product operations to achieve the attribute-hiding property. Extensive evaluations on real-world datasets demonstrate that our proposed schemes provide robust security guarantees while outperforming state-of-the-art solutions in both classification latency and bandwidth efficiency.
Chang-Ching Hung, Tsung-Che Hsieh, Yu-Chi Chen et al.· IEEE Transactions on Informa...· 0 citations
Private histogram computation is a fundamental building block for many data analytics tasks, enabling frequency analysis without revealing individual inputs. Existing protocols achieving robustness against malicious clients and servers typically require three servers with limited adversarial tolerance, restricting practicality.
In this work, we present Paras, the first two-server protocol for private histogram computation that achieves robustness against collusion between a malicious server and arbitrarily many malicious clients. Paras builds upon distributed point function-based approaches and introduces novel consistency checks leveraging vector oblivious linear evaluation (VOLE) to enforce both input correctness and output integrity. To realize these checks, we design two new cryptographic primitives: (1) aBV, an authenticated bit verification protocol that ensures VOLE committed shares correspond to valid bits, and (2) adIPA, an authenticated double inner product argument that enables secure consistency checks across two different VOLE sessions. These primitives may be of independent interest for other secure computation tasks.
We show that Paras is highly efficient and scalable: clients incur minimal cost independent of domain size, while servers achieve low per-client runtime, communication, and storage even at scale. For example, with 8192 clients over a domain of 128 inputs, each server requires only 14 ms runtime and 24 KB communication per client.
Dimitris Mouris, Lucas Piske, Pratik Sarkar et al.· IACR Communications in Crypt...· 0 citations
Cloud computing offers organizations scalable storage and computation, but outsourcing data processing to third-party infrastructure introduces serious privacy and confidentiality risks. Two complementary paradigms have emerged to address this challenge: homomorphic encryption (HE), which allows computation directly on encrypted data, and federated learning (FL), which enables collaborative model training without centralizing raw data. This paper presents a structured review of privacy-preserving data processing techniques for cloud environments built on HE and FL, individually and in hybrid combination. We propose a taxonomy of existing approaches, synthesize representative literature in a comparative table, illustrate a generic hybrid HE-FL architecture, and evaluate the two paradigms against criteria including data exposure, computational overhead, communication cost, resistance to inference attacks, and cloud deployment readiness. We further identify open challenges — including computational latency, key management, non-IID data distributions, and standardization gaps — and outline promising directions for future research, such as hardware-accelerated HE, adaptive encryption granularity, and standardized hybrid privacy frameworks for cloud-native machine learning.
Shivendra Shukla, Chandra Shekhar Gautam, Divyansh Tiwari· International Journal of Cre...· 0 citations