arXiv subjects

Bingsheng Zhang

Publications and source records attributed to Bingsheng Zhang.

12 recordsLinked to original sources

arXiv2026

From Profiling to Parameterization: Physics-Guided Acoustic Eavesdropping via Smartphone Accelerometers

We present LEAKFORGE, a device-agnostic framework that converts cross-device accelerometer eavesdropping into a physics-guided data-generation problem. Crucially, device-specific leakage is not arbitrary; its dominant variation lies within a constrained family of audio-to-accelerometer transfer functions. LEAKFORGE samples this family to synthesize large-scale, device-diverse accelerometer traces from ordinary speech, explicitly modeling electromechanical transfer, structural resonances, filtering, and aliasing. An eavesdropping model trained entirely in this synthetic domain can then be applied directly to traces from previously unseen smartphones.

cs.CR↗

arXiv2026

Dynamic Constraint Reconstruction Based Control Barrier Functions for Safety-Critical Control of High-Dimensional Manipulators

Control barrier functions (CBFs) provide formal safety guarantees for constrained nonlinear systems, but their effectiveness relies on accurate system dynamics. In high-dimensional manipulators subject to unknown disturbances and model uncertainties, fixed safety constraints constructed from nominal dynamics may become inconsistent with the actual system behavior, leading to safety degradation or excessive conservatism. This paper proposes a dynamic constraint reconstruction based control barrier function (DCR-CBF) framework for safety-critical control of disturbed robotic manipulators. An extended state observer is employed to estimate lumped disturbances online, and the estimated disturbance is incorporated into high-order control barrier functions to reconstruct safety constraints according to the estimated true dynamics. To address estimation inaccuracies, a safety margin is introduced, and a sufficient condition is derived to guarantee forward invariance under bounded estimation errors. Simulation studies on a 4-DOF excavation manipulator demonstrate that the proposed DCR-CBF method achieves zero safety violation under strong unknown disturbances while significantly improving trajectory-tracking performance compared with standard and robust CBF methods.

eess.SY↗

arXiv2025

Privacy-Preserving Inference for Quantized BERT Models

With the increasing deployment of generative machine learning models in privacy-sensitive domains such as healthcare and personalized services, ensuring secure inference has become a critical challenge. Secure multi-party computation (MPC) enables privacy-preserving model inference but suffers from high communication and computation overhead. The main bottleneck lies in the expensive secure evaluation of floating-point operations. Quantization offers a promising solution by converting floating-point operations into lower-precision integer computations, significantly reducing overhead. However, existing MPC-based quantized inference methods either rely on public quantization parameters-posing privacy risks-or suffer from inefficiencies, particularly in handling nonlinear functions such as activations and softmax. In this work, we propose a fine-grained, layer-wise quantization scheme and support 1-bit weight fully connected layers in a secure setting. We design a multi-input lookup table protocol to evaluate softmax efficiently and securely. Furthermore, we use dual secret sharing schemes and perform precision conversions via lookup tables, eliminating truncation overhead entirely. Experimental evaluation on BERT-base models demonstrates that our approach achieves up to $8\times$ speedup compared to Lu \emph{et al}. (NDSS 25), $9\times$ speedup compared to Gupta \emph{et al}. (PETS 24) and $22 \times$ speedup compared to Knott \emph{et al}. (NeurIPS 21).

cs.LG↗

arXiv2025

Accelerating Private Large Transformers Inference through Fine-grained Collaborative Computation

Homomorphic encryption (HE) and secret sharing (SS) enable computations on encrypted data, providing significant privacy benefits for large transformer-based models (TBM) in sensitive sectors like medicine and finance. However, private TBM inference incurs significant costs due to the coarse-grained application of HE and SS. We present FASTLMPI, a new approach to accelerate private TBM inference through fine-grained computation optimization. Specifically, through the fine-grained co-design of homomorphic encryption and secret sharing, FASTLMPI achieves efficient protocols for matrix multiplication, SoftMax, LayerNorm, and GeLU. In addition, FASTLMPI introduces a precise segmented approximation technique for differentiable non-linear, improving its fitting accuracy while maintaining a low polynomial degree. Compared to solution BOLT (S&P'24), FASTLMPI shows a remarkable 54% to 64% decrease in runtime and an impressive 72.2% reduction in communication costs.

cs.CR↗

arXiv2024

The Communication-Friendly Privacy-Preserving Machine Learning against Malicious Adversaries

With the increasing emphasis on privacy regulations, such as GDPR, protecting individual privacy and ensuring compliance have become critical concerns for both individuals and organizations. Privacy-preserving machine learning (PPML) is an innovative approach that allows for secure data analysis while safeguarding sensitive information. It enables organizations to extract valuable insights from data without compromising privacy. Secure multi-party computation (MPC) is a key tool in PPML, as it allows multiple parties to jointly compute functions without revealing their private inputs, making it essential in multi-server environments. We address the performance overhead of existing maliciously secure protocols, particularly in finite rings like $\mathbb{Z}_{2^\ell}$, by introducing an efficient protocol for secure linear function evaluation. We implement our maliciously secure MPC protocol on GPUs, significantly improving its efficiency and scalability. We extend the protocol to handle linear and non-linear layers, ensuring compatibility with a wide range of machine-learning models. Finally, we comprehensively evaluate machine learning models by integrating our protocol into the workflow, enabling secure and efficient inference across simple and complex models, such as convolutional neural networks (CNNs).

cs.CR↗

arXiv2021

UC Modelling and Security Analysis of the Estonian IVXV Internet Voting System

Estonian Internet voting has been used in national-wide elections since 2005. However, the system was initially designed in a heuristic manner, with very few proven security guarantees. The Estonian Internet voting system has constantly been evolving throughout the years, with the latest version (code-named IVXV) implemented in 2018. Nevertheless, to date, no formal security analysis of the system has been given. In this work, for the first time, we provide a rigorous security modeling for the Estonian IVXV system as a ceremony, attempting to capture the effect of actual human behavior on election verifiability in the universal composability (UC) framework. Based on the voter behavior statistics collected from three actual election events in Estonia, we show that IVXV achieves end-to-end verifiability in practice despite the fact that only $4\%$ (on average) of the Estonian voters audit their ballots.

cs.CR↗

arXiv2016

Distributed, End-to-end Verifiable, and Privacy-Preserving Internet Voting Systems

E-voting systems are a powerful technology for improving democracy. Unfortunately, prior voting systems have single points-of-failure, which may compromise availability, privacy, or integrity of the election results. We present the design, implementation, security analysis, and evaluation of the D-DEMOS suite of distributed, privacy-preserving, and end-to-end verifiable e-voting systems. We present two systems: one asynchronous and one with minimal timing assumptions but better performance. Our systems include a distributed vote collection subsystem that does not require cryptographic operations on behalf of the voter. We also include a distributed, replicated and fault-tolerant Bulletin Board component, that stores all necessary election-related information, and allows any party to read and verify the complete election process. Finally, we incorporate trustees, who control result production while guaranteeing privacy and end-to-end-verifiability as long as their strong majority is honest. Our suite of e-voting systems are the first whose voting operation is human verifiable, i.e., a voter can vote over the web, even when her web client stack is potentially unsafe, without sacrificing her privacy, and still be assured her vote was recorded as cast. Additionally, a voter can outsource election auditing to third parties, still without sacrificing privacy. We provide a model and security analysis of the systems, implement complete prototypes, measure their performance experimentally, and demonstrate their ability to handle large-scale elections. Finally, we demonstrate the performance trade-offs between the two versions of the system. A preliminary version of our system was used to conduct exit-polls at three voting sites for two national-level elections and is being adopted for use by the largest civil union of workers in Greece, consisting of over a half million members.

cs.CR↗

arXiv2015

D-DEMOS: A distributed, end-to-end verifiable, internet voting system

E-voting systems have emerged as a powerful technology for improving democracy by reducing election cost, increasing voter participation, and even allowing voters to directly verify the entire election procedure. Prior internet voting systems have single points of failure, which may result in the compromise of availability, voter secrecy, or integrity of the election results. In this paper, we present the design, implementation, security analysis, and evaluation of D-DEMOS, a complete e-voting system that is distributed, privacy-preserving and end-to-end verifiable. Our system includes a fully asynchronous vote collection subsystem that provides immediate assurance to the voter her vote was recorded as cast, without requiring cryptographic operations on behalf of the voter. We also include a distributed, replicated and fault-tolerant Bulletin Board component, that stores all necessary election-related information, and allows any party to read and verify the complete election process. Finally, we also incorporate trustees, i.e., individuals who control election result production while guaranteeing privacy and end-to-end-verifiability as long as their strong majority is honest. Our system is the first e-voting system whose voting operation is human verifiable, i.e., a voter can vote over the web, even when her web client stack is potentially unsafe, without sacrificing her privacy, and still be assured her vote was recorded as cast. Additionally, a voter can outsource election auditing to third parties, still without sacrificing privacy. Finally, as the number of auditors increases, the probability of election fraud going undetected is diminished exponentially. We provide a model and security analysis of the system. We implement a prototype of the complete system, we measure its performance experimentally, and we demonstrate its ability to handle large-scale elections.

cs.CR↗

arXiv2015

On the emergence of the Navier-Stokes-$α$ model for turbulent channel flows

In a series of papers (see \cite{CDT02} and the pertinent references therein) the 3D Navier-Stokes-$α$ model were shown to be a useful complement to the 3D Navier-Stokes equations; and in particular, to be a good Reynolds version of the latter equations. In this work, we introduce a simple Reynolds averaging which, due to the wall roughness, transforms the Navier-Stokes equations into the Navier-Stokes-$α$ model.

math.AP↗

arXiv2015

On solutions of the 2D Navier-Stokes equations with constant energy and enstrophy

It is not yet known if the global attractor of the space periodic 2D Navier-Stokes equations contains nonstationary solutions $u(x,t)$ such that their energy and enstrophy per unit mass are constant for every $t \in (-\infty, \infty)$. The study of the properties of such solutions was initiated in \cite{CMM13}, where, due to the hypothetical existence of such solutions, they were called "ghost solutions". In this work, we introduce and study geometric structures shared by all ghost solutions. This study led us to consider a subclass of ghost solutions for which those geometric structures have a supplementary stability property. In particular, we show that the wave vectors of the active modes of this subclass of ghost solutions must satisfy certain supplementary constraints. We also found a computational way to check for the existence of these ghost solutions.

math-ph↗

arXiv2013

On whether zero is in the global attractor of the 2D Navier-Stokes equations

The set of nonzero external forces for which the zero function is in the global attractor of the 2D Navier-Stokes equations is shown to be meagre in a Fréchet topology. A criterion in terms of a Taylor expansion in complex time is used to characterize the forces in this set. This leads to several relations between certain Gevrey subclasses of $C^{\infty}$ and a new upper bound for a Gevrey norm of solutions in the attractor, valid in the strip of analyticity in time.

math.AP↗

arXiv2013

Time analyticity with higher norm estimates for the 2D Navier-Stokes equations

This paper establishes bounds on norms of all orders for solutions on the global attractor of the 2D Navier-Stokes equations, complexified in time. Specifically, for periodic boundary conditions on $[0,L]^2$, and a force $g\in\calD(A^{\frac{α-1}{2}})$, we show there is a fixed strip about the real time axis on which a uniform bound $|A^αu|< m_ανκ_0^α$ holds for each $α\in \bN$. Here $ν$ is viscosity, $\k0=2π/L$, and $m_α$ is explicitly given in terms of $g$ and $α$. We show that if any element in $\calA$ is in $\D(A^α)$, then all of $\calA$ is in $\D(A^α)$, and likewise with $\D(A^α)$ replaced by $C^\infty(Ω)$. We demonstrate the universality of this "all for one, one for all" law on the union of a hierarchal set of function classes. Finally, we treat the question of whether the zero solution can be in the global attractor for a nonzero force by showing that if this is so, the force must be in a particular function class.

math.DS↗