Efficient benchmarking of algorithm configurators via model-based surrogates

The optimization of algorithm (hyper-)parameters is crucial for achieving peak performance across a wide range of domains, ranging from deep neural networks to solvers for hard combinatorial problems. However, the proper evaluation of new algorithm configuration (AC) procedures (or configurators) is hindered by two key hurdles. First, AC scenarios are hard to set up, including the target algorithm to be optimized and the problem instances to be solved. Second, and even more significantly, they are computationally expensive: a single configurator run involves many costly runs of the target algorithm. Here, we propose a benchmarking approach that uses surrogate scenarios, which are computationally cheap while remaining close to the original AC scenarios. These surrogate scenarios approximate the response surface corresponding to true target algorithm performance using a regression model. In our experiments, we construct and evaluate surrogate scenarios for hyperparameter optimization as well as for AC problems that involve performance optimization of solvers for hard combinatorial problems. We generalize previous work by building surrogates for AC scenarios with multiple problem instances, stochastic target algorithms and censored running time observations. We show that our surrogate scenarios capture overall important characteristics of the original AC scenarios from which they were derived, while being much easier to use and orders of magnitude cheaper to evaluate.     

A posteriori finite element bounds for linear-functional outputs of elliptic partial differential equations

We present a domain decomposition finite element technique for efficiently generating lower and upper bounds to outputs which are linear functionals of the solutions to symmetric or nonsymmetric second-order coercive linear partial differential equations in two space dimensions. The method is based upon the construction of an augmented Lagrangian, in which the objective is a quadratic ‘energy’ reformulation of the desired output, and the constraints are the finite element equilibrium equations and intersubdomain continuity requirements. The bounds on the output for a suitably fine ‘truth-mesh’ discretization are then derived by appealing to a dual max min relaxation evaluated for optimally chosen adjoint and hybrid-flux candidate Lagrange multipliers generated by a K-element coarser ‘working-mesh’ approximation. Independent of the form of the original partial differential equation, the computation on the truth mesh is reduced to K decoupled subdomain-local, symmetric Neumann problems. The technique is illustrated for the convection-diffusion and linear elasticity equations.     

Attenuated effects of chitosan-capped gold nanoparticles on LPS-induced toxicity in laboratory rats

The impact of nanoparticles in medicine and biology has increased rapidly in recent years. Gold nanoparticles (AuNP) have advantageous properties such as chemical stability, high electron density and affinity to biomolecules. However, the effects of AuNP on human body after repeated administration are still unclear. Therefore, the purpose of the present study was to evaluate the effects of gold-11.68 nm (AuNP1, 9.8 μg) and gold-22.22 nm (AuNP2, 19.7 μg) nanoparticles capped with chitosan on brain and liver tissue reactivity in male Wistar rats exposed to lipopolysaccharide (LPS from Escherichia coli serotype 0111:B4, 250 μg) upon 8 daily sessions of intraperitoneal administration. Our results suggest that the smaller size of chitosan-capped AuNP shows the protective effects against LPS-induced toxicity, suggesting a very high potential for biomedical applications.     

基于深度学习的事件抽取研究综述

事件抽取是从非结构化的自然语言文本中自动抽取用户感兴趣的事件信息, 并以结构化的形式表示出来. 事件抽取是自然语言处理与理解中的重要方向, 在政府公共事务管理、金融业务、生物医学等不同领域有着很高的应用价值. 根据对人工标注数据的依赖程度, 目前基于深度学习的事件抽取方法主要分为两类: 有监督和远程监督学习方法. 对当前深度学习中事件抽取技术进行了全面的综述. 围绕有监督中CNN、RNN、GAN、GCN与远程监督等方法, 系统地总结了近几年的研究情况, 并对不同的深度学习模型的性能进行了详细对比与分析. 最后, 对事件抽取面临的挑战进行了分析, 针对研究趋势进行了展望.     

Efficiently sampling nested Archimedean copulas

Efficient sampling algorithms for both Archimedean and nested Archimedean copulas are presented. First, efficient sampling algorithms for the nested Archimedean families of Ali-Mikhail-Haq, Frank, and Joe are introduced. Second, a general strategy how to build a nested Archimedean copula from a given Archimedean generator is presented. Sampling this copula involves sampling an exponentially tilted stable distribution. A fast rejection algorithm is developed for the more general class of tilted Archimedean generators. It is proven that this algorithm reduces the complexity of the standard rejection algorithm to logarithmic complexity. As an application it is shown that the fast rejection algorithm outperforms existing algorithms for sampling exponentially tilted stable distributions involved, e.g., in nested Clayton copulas. Third, with the additional help of randomization of generator parameters, explicit sampling algorithms for several nested Archimedean copulas based on different Archimedean families are found. Additional results include approximations and some dependence properties, such as Kendall’s tau and tail dependence parameters. The presented ideas may also apply in the more general context of sampling distributions given by their Laplace-Stieltjes transforms.     

Structural and electronic properties of tellurite glasses

The structural properties of some tellurite glasses were investigated by FT-IR spectroscopy, density measurements, and quantum chemical calculations. Main results reveal that the ratio TeO₄/TeO₃ is found to decrease in the order V₂O₅ > B₂O₃ > P₂O₅. For borate–tellurate glasses, the Van Hove singularities corresponding to Te 5s orbital-derived states are cleft suggesting that there are strong tellurium–oxygen interactions. On the other hand, a strong effect of TeO₂ on the vitreous B₂O₃ network is also demonstrated by FT-IR spectrum. This effect yields the apparition of small peaks in the region ranges between 800 and 1600 cm⁻¹ and probably the partial crystallization of the sample. Its spectral features are due to the B–O bond stretching of [BO₄] and [BO₃] structural units. The quantum chemical data obtained by us show that phosphate–tellurite and vanado–tellurate glasses can behave as semiconductors, whereas borate–tellurite glasses as insulators because the gap between the valence and conduction bands is >3 eV.     

Influence of a dry fractionation of butterfat on the content of fatty acids including conjugated linoleic acids

There is a growing demand among consumers for food products with natural nutritional–physiological advantages over comparable conventional products. As part of an EU project, a process using dry fractionation is evaluated that enables the targeted low-input enrichment of conjugated linoleic acids (CLA) in milk fat. Furthermore, the distribution of CLA isomers in the fat fractions was analysed. In the olein fraction for highland butter a CLA enrichment of 15.3% was obtained. The yield of the CLA rich olein fraction was 44.5% of the total amount of olein and stearin. There were significant increases during the first fractionation step of highland butter for the concentration of the CLA isomer cis-9, trans-11 (P ≤ 0.05) and during the second fractionation step for the concentration of CLA isomers cis-9, trans-11; trans-11, cis-13 (P ≤ 0.05) and trans-7, cis-9 (P ≤ 0.01). Experiments carried out demonstrate that the selected physical separation process enables CLA enrichment but the increase is too minor to achieve any decisive positive impact on human health and therefore too costly as an industrial CLA enrichment process.