Data poisoning attacks, where adversaries manipulate training data to degrade model performance, are an emerging threat as machine learning becomes widely deployed in sensitive applications. This paper provides a comp...
Data poisoning attacks, where adversaries manipulate training data to degrade model performance, are an emerging threat as machine learning becomes widely deployed in sensitive applications. This paper provides a comprehensive overview of data poisoning including attack techniques, adversary incentives, impacts on security and reliability, detection methods, defenses, and key research gaps. We examine label flipping, instance injection, backdoors, and other attack categories that enable malicious outcomes ranging from IP theft to accidents in autonomous systems. Promising detection approaches include statistical tests, robust learning, and forensics. However, significant challenges remain in translating academic defenses like adversarial training and sanitization into practical tools ready for operational use. With safety and trustworthiness at stake, more research on benchmarking evaluations, adaptive attacks, fundamental tradeoffs, and real-world deployment of defenses is urgently needed. Understanding vulnerabilities and developing resilient machine learning pipelines will only grow in importance as data integrity is fundamental to developing safe artificial intelligence.
The Lagrangian vortex particle method for two-dimensional flow simulation around airfoils is considered. Generally, the most time-consuming operation in the vortex method is vortex particles interaction simulation, es...
The Lagrangian vortex particle method for two-dimensional flow simulation around airfoils is considered. Generally, the most time-consuming operation in the vortex method is vortex particles interaction simulation, especially if it is performed directly 'point-to-point' for all pairs of particles. The result of such operation is vortex particles convective velocities. This problem is similar to the N-body problem and has squared computational complexity O(N2). The only efficient approach for its solution in a reasonable time for large N is to use approximate fast algorithms. The fast method suggested by Barnes and Hut having logarithmic computational complexity is considered for the mentioned problem of convective velocities computation. It also has been adapted for other operations in the vortex method such as a system of algebraic equations iterative solution and efficient right-hand side computation. As the result, the Barnes — Hut method allows reducing the whole time of calculations by several tens of times.
The problem of the accuracy improving is considered for vortex method. It is shown, that the airfoil surface line representation as a polygon consisting of rectilinear panels can lead to incorrect behavior of the nume...
The problem of the accuracy improving is considered for vortex method. It is shown, that the airfoil surface line representation as a polygon consisting of rectilinear panels can lead to incorrect behavior of the numerical solution of boundary integral equation with respect to unknown vortex sheet intensity, especially in case of considerably different lengths of the neighboring panels. The approach is developed which makes it possible to take into account explicitly the curvature of the surface line. The approximate formulae for the coefficients of linear system are derived which permit to obtain rather “good” numerical solution for arbitrary airfoil discretization.
The problem of numerical solution of the boundary integral equation arising in 2D Lagrangian meshless vortex particle methods is considered. The Viscous vortex domains method used for viscous flow simulation, that lea...
The problem of numerical solution of the boundary integral equation arising in 2D Lagrangian meshless vortex particle methods is considered. The Viscous vortex domains method used for viscous flow simulation, that leads to a large number of vortex particles in the flow domain, especially in the near-wall region. From a mathematical point of view, it leads, in turn, to the fast oscillating right-hand side of the integral equation. An original correction technique is developed that permits one to obtain a numerical solution with high resolution without grid refining. It is based on the Galerkin approach where additional terms are included in numerical solution expansion, that correspond to the subgrid-scale of the solution variation.
The algorithm of the Viscous Vortex Domain method implemented in the author's code VM2D is investigated. This code allows simulating 2D viscous incompressible flows and solving fluid-structure interaction problems...
The algorithm of the Viscous Vortex Domain method implemented in the author's code VM2D is investigated. This code allows simulating 2D viscous incompressible flows and solving fluid-structure interaction problems. The verification of the algorithm, namely, the correctness of taking into account the viscosity effect, is performed on well-studied test problems of the flow simulation around a circular cylinder for Reynolds numbers in the range 20 . . . 200: the problem of the impulse start of the circular cylinder is considered, and the dependency of the flow separation angle on the Reynolds number is investigated. The comparison shows that for low Reynolds numbers (up to 140), the flow separation angle obtained in numerical simulations by using the VM2D code is in good agreement with the results of other studies. For higher Reynolds numbers, the estimated separation angle is slightly overestimated.
A droplet impact on a deep pool can induce macro-scale or micro-scale effects like a crown splash, a high-speed jet, formation of secondary droplets or thin liquid films, etc. It depends on the diameter and velocity o...
A droplet impact on a deep pool can induce macro-scale or micro-scale effects like a crown splash, a high-speed jet, formation of secondary droplets or thin liquid films, etc. It depends on the diameter and velocity of the droplet, liquid properties, effects of external forces and other factors that a ratio of dimensionless criteria can account for. In the present research, we considered the droplet and the pool consist of the same viscous incompressible liquid. We took surface tension into account but neglected gravity forces. We used two open-source codes (OpenFOAM and Gerris) for our computations. We review the possibility of using these codes for simulation of processes in free-surface flows that may take place after a droplet impact on the pool. Both codes simulated several modes of droplet impact. We estimated the effect of liquid properties with respect to the Reynolds number and Weber number. Numerical simulation enabled us to find boundaries between different modes of droplet impact on a deep pool and to plot corresponding mode maps. The ratio of liquid density to that of the surrounding gas induces several changes in mode maps. Increasing this density ratio suppresses the crown splash.
system Management Mode (SMM) is the highest-privileged operating mode of x86 and x86-64 processors. Through SMM exploitation, attackers can tamper with the Unified Extensible Firmware Interface (UEFI) firmware, disabl...
详细信息
A novel algorithm of the Lagrangian vortex method is considered for incompressible flow simulation. The boundary condition on the body surface is satisfied by vortex sheet introduction, which intensity is determined f...
A novel algorithm of the Lagrangian vortex method is considered for incompressible flow simulation. The boundary condition on the body surface is satisfied by vortex sheet introduction, which intensity is determined from the equality between the tangential components of flow velocity limit value and the body surface velocity. High accuracy of velocity field reconstruction in the neighbourhood of the body surface permits to estimate unsteady aerodynamic loads for the complex-shaped bodies, including added masses tensor components with rather low computational cost.
Internal and inertial wave attractors appear as a result of propagation of waves obeying very specific dispersion relation. In closed containers, some boundaries of which are inclined with respect to the anisotropy di...
Internal and inertial wave attractors appear as a result of propagation of waves obeying very specific dispersion relation. In closed containers, some boundaries of which are inclined with respect to the anisotropy direction, this may result in domination of focusing of waves upon reflection, and after multiple reflections from the boundaries all the wave packets may be attracted by a closed loop or limit trajectory in physical space. It is well known that for wave attractors in plane geometries the change of the direction of anisotropy to the opposite will not change the geometry of the attractor and direction of flow of energy (in case of internal waves change of anisotropy direction means change of gravity direction and stratification growth). More complicated situation can arise for inertial wave attractors in fully 3D geometry. If we consider an annular layer with an inclined boundary, the change of the angle of inclination of a boundary to the opposite may also affect the geometry of the attractor, and the instability growth for higher amplitude of external forcing.
We consider the methodology of numerical schemes development for two-dimensional vortex method. We describe two different approaches to deriving integral equation for unknown vortex sheet intensity. We simulate the ve...
We consider the methodology of numerical schemes development for two-dimensional vortex method. We describe two different approaches to deriving integral equation for unknown vortex sheet intensity. We simulate the velocity of the surface line of an airfoil as the influence of attached vortex and source sheets. We consider a polygonal approximation of the airfoil and assume intensity distributions of free and attached vortex sheets and attached source sheet to be approximated with piecewise constant or piecewise linear (continuous or discontinuous) functions. We describe several specific numerical schemes that provide different accuracy and have a different computational cost. The study shows that a Galerkin-type approach to solving boundary integral equation requires computing several integrals and double integrals over the panels. We obtain exact analytical formulae for all the necessary integrals, which makes it possible to raise significantly the accuracy of vortex sheet intensity computation and improve the quality of velocity and vorticity field representation, especially in proximity to the surface line of the airfoil. All the formulae are written down in the invariant form and depend only on the geometric relationship between the positions of the beginnings and ends of the panels.
暂无评论