Добірка наукової літератури з теми "Maximum Worst Case Entropy Selector"

In this paper a probabilistic feedback technique to maximize the throughput of a generic True Random Bit Generator (TRBG) circuit, under a given constraint on the entropy, is discussed. In the proposed solution, the throughput of the device is dynamically and adaptively varied by an on-line entropy detector, such to obtain, with an arbitrary confidence level, an entropy greater than a given worst-case value. The approach, which has a general validity, introduces a method for making maximum use of the TRBG random bit generation capabilities, maximizing the generation throughput while preserving its entropy. It is different from the classical "open loop" TRBG design approach, in which the circuit parameter variability determines an uncertainty about the actual entropy of the device, with the proposed techniques the TRBG generation speed is varied under a given constraint on the entropy. The method can be applied to all those integrated TRBG circuits proposed in the literature and based on the uniform sampling of, e.g., random physical processes or chaotic dynamical systems.

For stability and reliability of real-world applications, the robustness of DNNs in unimodal tasks has been evaluated. However, few studies consider abnormal situations that a visual question answering (VQA) model might encounter at test time after deployment in the real-world. In this study, we evaluate the robustness of state-of-the-art VQA models to five different anomalies, including worst-case scenarios, the most frequent scenarios, and the current limitation of VQA models. Different from the results in unimodal tasks, the maximum confidence of answers in VQA models cannot detect anomalous inputs, and post-training of the outputs, such as outlier exposure, is ineffective for VQA models. Thus, we propose an attention-based method, which uses confidence of reasoning between input images and questions and shows much more promising results than the previous methods in unimodal tasks. In addition, we show that a maximum entropy regularization of attention networks can significantly improve the attention-based anomaly detection of the VQA models. Thanks to the simplicity, attention-based anomaly detection and the regularization are model-agnostic methods, which can be used for various cross-modal attentions in the state-of-the-art VQA models. The results imply that cross-modal attention in VQA is important to improve not only VQA accuracy, but also the robustness to various anomalies.

Indonesia has a large number of primate diversity where a majority of the species are threatened. In addition, climate change is conservation issues that biodiversity may likely face in the future, particularly among primates. Thus, species-distribution modeling was useful for conservation planning. Herein, we present protected areas (PA) recommendations with high nature-conservation importance based on species-richness changes. We performed maximum entropy (Maxent) to retrieve species distribution of 51 primate species across Indonesia. We calculated species-richness change and range shifts to determine the priority of PA for primates under mitigation and worst-case scenarios by 2050. The results suggest that the models have an excellent performance based on seven different metrics. Current primate distributions occupied 65% of terrestrial landscape. However, our results indicate that 30 species of primates in Indonesia are likely to be extinct by 2050. Future primate species richness would be also expected to decline with the alpha diversity ranging from one to four species per 1 km2. Based on our results, we recommend 54 and 27 PA in Indonesia to be considered as the habitat-restoration priority and refugia, respectively. We conclude that species-distribution modeling approach along with the categorical species richness is effectively applicable for assessing primate biodiversity patterns.

Sensitivity of output of a linear operator to its input can be quantified in various ways. In Control Theory, the input is usually interpreted as disturbance and the output is to be minimized in some sense. In stochastic worst-case design settings, the disturbance is considered random with imprecisely known probability distribution. The prior set of probability measures can be chosen so as to quantify how far the disturbance deviates from the white-noise hypothesis of Linear Quadratic Gaussian control. Such deviation can be measured by the minimal Kullback-Leibler informational divergence from the Gaussian distributions with zero mean and scalar covariance matrices. The resulting anisotropy functional is defined for finite power random vectors. Originally, anisotropy was introduced for directionally generic random vectors as the relative entropy of the normalized vector with respect to the uniform distribution on the unit sphere. The associated a-anisotropic norm of a matrix is then its maximum root mean square or average energy gain with respect to finite power or directionally generic inputs whose anisotropy is bounded above by a≥0. We give a systematic comparison of the anisotropy functionals and the associated norms. These are considered for unboundedly growing fragments of homogeneous Gaussian random fields on multidimensional integer lattice to yield mean anisotropy. Correspondingly, the anisotropic norms of finite matrices are extended to bounded linear translation invariant operators over such fields.

This article presents a probabilistic approach to investigate the effect of parametric uncertainties on the mean power, tip deflection, and tip velocity of linear and nonlinear energy harvesting systems. Recently developed conjugate unscented transformation algorithm is used to compute the statistical moments of the output variables with multidimensional Gaussian uncertainty in parameters. The principle of maximum entropy is used to construct the probability density function of output variables from the knowledge of obtained statistical moments. The probability density functions for mean power were significantly complicated in shape with two and three distinct peaks for the nonlinear monostable and nonlinear bistable harvesters, respectively. Monte-Carlo simulations with N = 8 × 104 samples for monostable harvester and N = 6.5 × 104 samples for bistable harvester were used for validating the probability density functions. It is concluded that conjugate unscented transformation methodology affords a significant computational advantage without compromising accuracy. In addition, using conjugate unscented transformation method, we show that the dependence of mean power on parameters (excitation frequency, excitation amplitude, etc.), when multidimensional uncertainties are present, is decidedly different relative to a purely deterministic trend. The discrepancy in predicted power between the deterministic and uncertain trends for the monostable harvester, for instance, reach a maximum of 100%, 234%, and 110% for base frequency, base acceleration, and magnet gap, respectively. The deterministic trend consistently overestimates the harvested power relative to the uncertain trends. This work, therefore, may have applications in evaluating “worst case scenario” for harvested power. The major advantage of the presented methodology relative to extant techniques in energy harvesting literature is the accurate and computationally effective applicability to multidimensional uncertainty in parameters.

Gully erosion is a problem; therefore, it must be predicted using highly accurate predictive models to avoid losses caused by gully development and to guarantee sustainable development. This research investigates the predictive performance of seven multiple-criteria decision-making (MCDM), statistical, and machine learning (ML)-based models and their ensembles for gully erosion susceptibility mapping (GESM). A case study of the Dasjard River watershed, Iran uses a database of 306 gully head cuts and 15 conditioning factors. The database was divided 70:30 to train and verify the models. Their performance was assessed with the area under prediction rate curve (AUPRC), the area under success rate curve (AUSRC), accuracy, and kappa. Results show that slope is key to gully formation. The maximum entropy (ME) ML model has the best performance (AUSRC = 0.947, AUPRC = 0.948, accuracy = 0.849 and kappa = 0.699). The second best is the random forest (RF) model (AUSRC = 0.965, AUPRC = 0.932, accuracy = 0.812 and kappa = 0.624). By contrast, the TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) model was the least effective (AUSRC = 0.871, AUPRC = 0.867, accuracy = 0.758 and kappa = 0.516). RF increased the performance of statistical index (SI) and frequency ratio (FR) statistical models. Furthermore, the combination of a generalized linear model (GLM), and functional data analysis (FDA) improved their performances. The results demonstrate that a combination of geographic information systems (GIS) with remote sensing (RS)-based ML models can successfully map gully erosion susceptibility, particularly in low-income and developing regions. This method can aid the analyses and decisions of natural resources managers and local planners to reduce damages by focusing attention and resources on areas prone to the worst and most damaging gully erosion.

In recent years, cybersecurity is gaining more and more importance. Cryptography is used in numerous applications, such as authentication and encryption of data in communications, access control to restricted or protected areas, electronic payments. It is safe to assume that the presence of cryptographic systems in future technologies will become increasingly pervasive, leading to a greater demand for energy efficiency, hardware reliability, integration, portability, and security. However, this pervasiveness introduces new challenges: the implementation of conventional cryptographic standards approved by NIST requires the achievement of performance in terms of timing, chip area, power and resource consumption that are not compatible with reduced complexity hardware devices, such as IoT systems. In response to this limitation, lightweight cryptography comes into play - a branch of cryptography that provides tailor-made solutions for resource-limited devices. One of the fundamental classes of cryptographic hardware primitives is represented by Random Number Generators (RNGs), that is, systems that provide sequences of integers that are supposed to be unpredictable. The circuits and systems that implement RNGs can be divided into two categories, namely Pseudo Random Number Generators (PRNGs) and True Random Number Generators (TRNGs). PRNGs are deterministic and possibly periodic finite state machines, capable of generating sequences that appear to be random. In other words, a PRNG is a device that generates and repeats a finite random sequence, saved in memory, or generated by calculation. A TRNG, on the other hand, is a device that generates random numbers based on real stochastic physical processes. Typically, a hardware TRNG consists of a mixed-signal circuit that is classified according to the stochastic process on which it is based. Specifically, the most used sources of randomness are chaotic circuits, high jitter oscillators, circuits that measure other stochastic processes. A chaotic circuit is an analog or mixed-signal circuit in which currents and voltages vary over time based on certain mathematical properties. The evolution over time of these currents and voltages can be interpreted as the evolution of the state of a chaotic nonlinear dynamical system. Jitter noise can instead be defined as the deviation of the output signal of an oscillator from its true periodicity, which causes uncertainty in its low-high and high-low transition times. Other possible stochastic processes that a TRNG can use may involve radioactive decay, photon detection, or electronic noise in semiconductor devices. TRNG proposals presented in the literature are typically designed in the form of Application Specific Integrated Circuits (ASICs). On the other hand, in recent years more and more researchers are exploring the possibility of designing TRNGs in Programmable Logic Devices (PLDs). A PLD offers, compared to an ASIC, clear advantages in terms of cost and versatility. At the same time, however, there is currently a widespread lack of trust in these PLD-based architectures, particularly due to strong cryptographic weaknesses found in Ring Oscillator-based solutions. The goal of this thesis is to show how this mistrust does not depend on poor performance in cryptographic terms of solutions for the generation of random numbers based on programmable digital technologies, but rather on a still immature approach in the study of TRNG architectures designed on PLDs. During the thesis chapters a new class of nonlinear circuits based on digital hardware is introduced that can be used as entropy sources for TRNGs implemented in PLDs, identified by the denomination of Digital Nonlinear Oscillators (DNOs). In Chapter 2 a novel class of circuits that can be used to design entropy sources for True Random Number Generation, called Digital Nonlinear Oscillators (DNOs), is introduced. DNOs constitute nonlinear dynamical systems capable of supporting complex dynamics in the time-continuous domain, although they are based on purely digital hardware. By virtue of this characteristic, these circuits are suitable for their implementation on Programmable Logic Devices. By focusing the analysis on Digital Nonlinear Oscillators implemented in FPGAs, a preliminary comparison is proposed between three different circuit topologies referable to the introduced class, to demonstrate how circuits of this type can have different characteristics, depending on their dynamical behavior and the hardware implementation. In Chapter 3 a methodology for the analysis and design of Digital Nonlinear Oscillators based on the evaluation of their electronics aspects, their dynamical behavior, and the information they can generate is formalized. The presented methodology makes use of different tools, such as figures of merit, simplified dynamical models, advanced numerical simulations and experimental tests carried out through implementation on FPGA. Each of these tools is analyzed both in its theoretical premises and through explanatory examples. In Chapter 4 the analysis and design methodologies of Digital Nonlinear Oscillators formalized in Chapter 3 are used to describe the complete workflow followed for the design of a novel DNO topology. This DNO is characterized by chaotic dynamical behaviors and can achieve high performance in terms of generated entropy, downstream of a reduced hardware complexity and high sampling frequencies. By exploiting the simplified dynamical model, the advanced numerical simulations in Cadence Virtuoso and the FPGA implementation, the presented topology is extensively analyzed both from a theoretical point of view (notable circuit sub-elements that make up the topology, bifurcation diagrams, internal periodicities) and from an experimental point of view (generated entropy, source autocorrelation, sensitivity to routing, application of standard statistical tests). In Chapter 5 an algorithm, called Maximum Worst-Case Entropy Selector (MWCES), that aims to identify, within a set of entropy sources, which offers the best performance in terms of worst-case entropy, also known in literature as "min-entropy", is presented. This algorithm is designed to be implemented in low-complexity digital architectures, suitable for lightweight cryptographic applications, thus allowing online maximization of the performance of a random number generation system based on Digital Nonlinear Oscillators. This chapter presents the theoretical premises underlying the algorithm formulation, some notable examples of its generic application and, finally, considerations related to its hardware implementation in FPGA.