A fully supervised universal adversarial perturbations and the progressive optimization

Universal Adversarial Perturbations(UAPs), which are image-agnostic adversarial perturbations, have been demonstrated to successfully deceive computer vision models. Proposed UAPs in the case of data-dependent, use the internal layers’ activation or the output layer’s decision values as supervision. In this paper, we use both of them to drive the supervised learning of UAP, termed as fully supervised UAP(FS-UAP), and design a progressive optimization strategy to solve the FS-UAP. Specifically, we define an internal layers supervised objective relying on multiple major internal layers’ activation to estimate the deviations of adversarial examples from legitimate examples. We also define an output layer supervised objective relying on the logits of output layer to evaluate attacking degrees. In addition, we use the UAP found by previous stage as the initial solution of the next stage so as to progressively optimize the UAP stage-wise. We use seven networks and ImageNet dataset to evaluate the proposed FS-UAP, and provide an in-depth analysis for the latent factors affecting the performance of universal attacks. The experimental results show that our FS-UAP (i) has powerful capability of cheating CNNs (ii) has superior transfer-ability across models and weak data-dependent (iii) is appropriate for both untarget and target attacks.

Improved Well-Control Progressive Optimization with Generalized Barycentric Coordinates and Manifold Mapping

Summary Well-control management is nowadays frequently approached by means of mathematical optimization. However, in many practical situations the optimization algorithms used are still computationally expensive. In this paper, we present progressive optimization (PO), a simulator-nonintrusivefour-stage methodology to accelerate optimal search substantially in well-controlapplications. The first stage of PO comprises a global exploration of the search space using design of experiments (DOEs). Thereafter, in the second stage, a fast-to-evaluate proxy model is constructed with the points considered in the experimental design. This proxy is based on generalized barycentric coordinates (GBCs), a generalization of the concept of barycentric coordinates used within a triangle. GBCs can be especially suited to problems in which nonlinearities are not strong, as is the case often for well-control optimization. This fact is supported by the good performance in these types of optimization problems of techniques that rely strongly on linearity assumptions, such as trajectory piecewise linearization, a procedure that is not always applicable due to its simulator-intrusive nature. In the third stage, the precision of the proxy model is iteratively improved and the enhanced surrogate model is reoptimized by means of manifold mapping (MM), a method that combines models with different levels of accuracy. MM has solid theoretical foundations and leads to efficient optimization schemes in multiple engineering disciplines. The final and fourth stage aims at additional improvement, resorting to direct optimization of the best solution from the previous stages. Nonlinear (operational) constraints are handled in PO with the filter method. The optimal search may be finalized earlier than at the fourth stage whenever the solution obtained is of satisfactory quality. PO is tested on two waterflooding problems built upon a synthetic model previously studied in well-control optimization literature. In these problems, which have 120 and 40 well controls and include nonlinear constraints, we observe for PO reductions in computational cost, for solutions of comparable quality, of approximately 30% and 50% with respect to Hooke-Jeeves direct search (HJDS), which, in turn, outperforms particle swarm optimization (PSO). HJDS and PSO are simulator-nonintrusive algorithms that usually perform well in optimization for oilfield operations. The novel concepts of GBC and MM within the framework of the PO paradigm can be extremely helpful for practitioners to efficiently deal with optimized well-control management. Savings of 50% in computing cost may be translated in practice into days of computations for just a single field and optimization run.

Optimization of road speed-sectioning by assessing the impact of a road speed limitation sign

Energy consumed by road vehicles has a high impact on climate changes; indeed this energy use accounts for 23% of total energy-related Green House Gases (GHG) emissions of 2014 global GHG emissions. GHG emissions are growing constantly year after year, in spite of global objectives (COP) and researches on vehicle efficiency and modal shift. The contribution of the infrastructure to lower this energy is less studied, since it is often seen as immuable or too costly. This paper aims to demonstrate that simple and low-cost solutions exist for that purpose. Particularly a methodology has been developed, based on an optimization of the speed layout over an itinerary in order to improve the eco- driving potential of a given road infrastructure. The key point of this work is that inconsistency often exists between vehicle dynamics, road longitudinal profile and changes in regulation speeds. These changes in speed are defining the speed- sectioning of a route, and an optimization of this speed-sectioning can be easily carried out while displacing or modifying speed signs. The objective of this study is to build an optimized speed sectioning which minimizes the fuel consumption for realistic traffic and various driver behaviors, while maintaining the required safety levels. A progressive optimization loop has been worked out with a Python script including an embedded microscopic road traffic simulator. As a result, an optimized speed-sectioning is leading to a gain of 227 ml for 60 minutes of simulated flow of 100 veh/h/lane, for a modification of a single speed changing point. The overall benefits are reduced energy consumption, air pollution and noise which otherwise would have been produced by braking. This work brings an effective optimization tool for road managers and its practical application is passive and inexpensive. This methodology is suitable for rural and urbanized territories and easily adaptable to any type of traffic in various countries. In perspectives, the optimization process could be extended to a full road route and to a wide range of different speed-sectioning layouts.

Stable Portfolio Selection Strategy for Mean-Variance-CVaR Model under High-Dimensional Scenarios

This paper aims to study stable portfolios with mean-variance-CVaR criteria for high-dimensional data. Combining different estimators of covariance matrix, computational methods of CVaR, and regularization methods, we construct five progressive optimization problems with short selling allowed. The impacts of different methods on out-of-sample performance of portfolios are compared. Results show that the optimization model with well-conditioned and sparse covariance estimator, quantile regression computational method for CVaR, and reweighted L1 norm performs best, which serves for stabilizing the out-of-sample performance of the solution and also encourages a sparse portfolio.

Design of a High-Performance Digital Slit-Lamp Microscope with Five-Switched Zoom

In this study, we design and present a five-fold digital slit-lamp microscope (DSLM) with built-in photographic lens and CCD. The initial structures of the front objective, Galilean telescope system, and photographic lens are systematically investigated and discussed in the design. A progressive optimization process is employed in the non-coaxial system design after the coaxial system achieves high performance. The analysis of spot diagrams and the modulation transfer function (MTF) show that this DSLM optical system achieves quasi-diffraction-limited performance and enables high-quality imaging for ophthalmic examination. Furthermore, tolerance analysis of this optical system is also performed, which provides a theoretical basis for machining and assembly. This design provides an idea for the design of a digital-zoom microscope in biomedical imaging instruments.