Instance Space Research Articles

Adaptive configuration of radial basis function network by regression tree allied with hybrid particle swarm optimization algorithm

Configuration of a radial basis function network (RBFN) comprises identifying the network parameters (inputs, centers as well as widths in RBF units, and weights between the hidden and output layers) and architecture. The issues of overfitting and local optima often happened during RBFN training. To rectify this situation, regression tree (RT), allied with hybrid particle swarm optimization (PSO) algorithm, was invoked to configure an RBFN to form the HPSORTRBFN algorithm in the present study. Discrete PSO was invoked to obtain an RT of the right size. The regions in the instance space defined by the leaf nodes of the grown RT were transformed into the centers in RBF units and the number of leaf nodes acted as the network structure. The splitting variables in RT became the inputs in RBFN. The widths and weights in RBFN were simultaneously optimized by continuous PSO. HPSORTRBFN was applied to predict the anti-HIV activities of 1-[(2-Hydroxyethoxy)methyl]-6-(phenylthio)thymine (HEPT) analogues and the bioactivities of flavonoid derivatives. The results showed RT allied with HPSO is able to configure a globally optimal RBFN and HPSORTRBFN owns superior modeling performance to RBFN and RT.

Electromagnetic wave scattering from a rough interface above a chiral medium: generalized field transforms

In order to derive full-wave solutions for electromagnetic wave scattering from rough interfaces between achiral media (free space for instance) and chiral media that satisfy generalized constitutive relations, it is necessary to employ complete modal expansions for the electromagnetic fields above and below the interface. To this end, the familiar Fourier transforms of the fields are expressed as generalized field transforms consisting of the radiation term, the lateral waves, and the surface waves. Maxwell's equations are converted into generalized telegraphists' equations [in the companion paper (this issue), J. Opt. Soc. Am. A 30, 335 (2013)] upon the imposition of exact boundary conditions. These telegraphists' equations are coupled first-order differential equations for the forward- and backward-traveling wave amplitudes associated with all the different species of waves (radiation, lateral, and surface waves) excited at the surface of the chiral medium. The analysis presented here includes the completeness and orthogonal relations of the basis functions associated with the modal expansions. This work is used to distinguish between depolarization due to the chiral properties of the medium and depolarization due to surface irregularities. It has applications in remote sensing and identification of biological and chemical materials based on their optical activity.

A probabilistic kernel approach for solving the multi-instance learning problems with different assumptions

Multi-instance learning (MIL) has received more and more attentions in the machine learning research field due to its theoretical interest and its applicability to diverse real-world problems. In this paper, we present a probabilistic kernel approach for the multi-instance learning problems with various multi-instance assumptions by imposing Gaussian process prior on an unobservable latent function defined on the instance space. Because the relationship between the bag and its instances, triggered by the multi-instance assumption, can be exactly captured by defining the likelihood function, we can deal with different multi-instance assumptions by employing different likelihood functions. Experimental results on several multi-instance problems show that the proposed algorithms are valid and can achieve superior performance to the published MIL algorithms.

난이도-거리 상관관계 기반의 문제 인스턴스 공간 분석

문제 인스턴스 탐색 혹은 자동 생성은 알고리즘 분석 및 테스트에 적용될 수 있으며, 하드웨어, 소프트웨어 프로그램, 계산 이론 등 다양한 수준에서 연구되어온 주제이다. 본 연구에서는 해(解) 공간에 사용된 목적값-거리 상관관계 분석을 문제 인스턴스 공간에 적용하였다. 문제 인스턴스의 목적값은 문제에 따라 알고리즘의 수행 시간과 최적해를 잘 구하는 정도로 정의하였다. 이러한 정의는 문제 인스턴스의 난이도로 해석할 수 있다. 상관관계는 3가지 측면에서 분석하였다: 첫째, 알고리즘과 거리 함수에 따른 상관관계 차이, 둘째, 알고리즘의 개선 전/후의 상관관계 변화, 셋째, 문제 인스턴스 공간과 해당 문제의 해 공간 사이의 연관성. 본 논문은 문제 인스턴스 공간에 상관계수 분석이 어떻게 적용될 수 있는지 보여주며, 문제 인스턴스 공간 분석을 본격적으로 다루는 첫번째 시도이다. Finding or automatically generating problem instance is useful for algorithm analysis/test. The topic has been of interest in the field of hardware/software engineering and theory of computation. We apply objective value-distance correlation analysis to problem spaces, as previous researchers applied it to solution spaces. According to problems, we define the objective function by (1) execution time of tested algorithm or (2) its optimality; this definition is interpreted as difficulty of the problem instance being solved. Our correlation analysis is based on the following aspects: (1) change of correlation when we use different algorithms or different distance functions for the same problem, (2) change of that when we improve the tested algorithm, (3) relation between a problem instance space and the solution space for the same problem. Our research demonstrates the way of problem instance space analysis and will accelerate the problem instance space analysis as an initiative research.

Analysis of convergence performance of neural networks ranking algorithm

The ranking problem is to learn a real-valued function which gives rise to a ranking over an instance space, which has gained much attention in machine learning in recent years. This article gives analysis of the convergence performance of neural networks ranking algorithm by means of the given samples and approximation property of neural networks. The upper bounds of convergence rate provided by our results can be considerably tight and independent of the dimension of input space when the target function satisfies some smooth condition. The obtained results imply that neural networks are able to adapt to ranking function in the instance space. Hence the obtained results are able to circumvent the curse of dimensionality on some smooth condition.

Hybrid probabilities and error-domain structural identification using ambient vibration monitoring

For the assessment of structural behaviour, many approaches are available to compare model predictions with measurements. However, few approaches include uncertainties along with dependencies associated with models and observations. In this paper, an error-domain structural identification approach is proposed using ambient vibration monitoring (AVM) as the input. This approach is based on the principle that in science, data cannot truly validate a hypothesis, it can only be used to falsity it. Error-domain model falsification generates a space of possible model instances (combination of parameters), obtains predictions for each of them and then rejects instances that have unlikely differences (residuals) between predictions and measurements. Models are filtered in a two step process. Firstly a comparison of mode shapes based on MAC criterion ensures that the same modes are compared. Secondly, the frequencies from each model instance are compared with the measurements. The instances for which the difference between the predicted and measured value lie outside threshold bounds are discarded. In order to include “uncertainty of uncertainty” in the identification process, a hybrid probability scheme is also presented. The approach is used for the identification of the Langensand Bridge in Switzerland. It is used to falsify the hypothesis that the bridge was behaving as designed when subjected to ambient vibration inputs, before opening to the traffic. Such small amplitudes may be affected by low-level bearing-device friction. This inadvertently increased the apparent stiffness of the structure by 17%. This observation supports the premiss that ambient vibration surveys should be cross-checked with other information sources, such as numerical models, in order to avoid misinterpreting the data.

Breast Ultrasound Image Classification Based on Multiple-Instance Learning

Breast ultrasound (BUS) image segmentation is a very difficult task due to poor image quality and speckle noise. In this paper, local features extracted from roughly segmented regions of interest (ROIs) are used to describe breast tumors. The roughly segmented ROI is viewed as a bag. And subregions of the ROI are considered as the instances of the bag. Multiple-instance learning (MIL) method is more suitable for classifying breast tumors using BUS images. However, due to the complexity of BUS images, traditional MIL method is not applicable. In this paper, a novel MIL method is proposed for solving such task. First, a self-organizing map is used to map the instance space to the concept space. Then, we use the distribution of the instances of each bag in the concept space to construct the bag feature vector. Finally, a support vector machine is employed for classifying the tumors. The experimental results show that the proposed method can achieve better performance: the accuracy is 0.9107 and the area under receiver operator characteristic curve is 0.96 (p < 0.005).

Bayesian multi-instance multi-label learning using Gaussian process prior

Multi-instance multi-label learning (MIML) is a newly proposed framework, in which the multi-label problems are investigated by representing each sample with multiple feature vectors named instances. In this framework, the multi-label learning task becomes to learn a many-to-many relationship, and it also offers a possibility for explaining why a concerned sample has the certain class labels. The connections between instances and labels as well as the correlations among labels are equally crucial information for MIML. However, the existing MIML algorithms can rarely exploit them simultaneously. In this paper, a new MIML algorithm is proposed based on Gaussian process. The basic idea is to suppose a latent function with Gaussian process prior in the instance space for each label and infer the predictive probability of labels by integrating over uncertainties in these functions using the Bayesian approach, so that the connection between instances and every label can be exploited by defining a likelihood function and the correlations among labels can be identified by the covariance matrix of the latent functions. Moreover, since different relationships between instances and labels can be captured by defining different likelihood functions, the algorithm may be used to deal with the problems with various multi-instance assumptions. Experimental results on several benchmark data sets show that the proposed algorithm is valid and can achieve superior performance to the existing ones.

Relevance vector machine based infinite decision agent ensemble learning for credit risk analysis

In this paper, a relevance vector machine based infinite decision agent ensemble learning (RVM Ideal ) system is proposed for the robust credit risk analysis. In the first level of our model, we adopt soft margin boosting to overcome overfitting. In the second level, the RVM algorithm is revised for boosting so that different RVM agents can be generated from the updated instance space of the data. In the third level, the perceptron Kernel is employed in RVM to simulate infinite subagents. Our system RVM Ideal also shares some good properties, such as good generalization performance, immunity to overfitting and predicting the distance to default. According to the experimental results, our proposed system can achieve better performance in term of sensitivity, specificity and overall accuracy.

Learning to Rank with Bayesian Evidence Framework

The problem of ranking has recently gained attention in data learning. The goal ranking is to learn a real-valued ranking function that induces a ranking or ordering over an instance space. In this paper, we apply popular Bayesian techniques on ranking support vector machine. We propose a novel differentiable loss function called trigonometric loss function with the desirable characteristic of natural normalization in the likelihood function, and then follow standard Gaussian processes techniques to set up a Bayesian framework. In this framework, Bayesian inference is used to implement model adaptation, while keeping the merits of ranking SVM. Experimental results on data sets indicate the usefulness of this approach.

Optimal Solutions for Mixed Bundling of Two Independently-Valued Goods

This paper develops analytical results and insights for the mixed bundling problem of pricing a product line consisting of two component goods and a bundle of the two goods. Consumer valuations for the two goods follow a symmetric bivariate uniform distribution, a setting which has previously been considered analytically intractable. Prior analytical work has demonstrated the superiority of mixed bundling, but failed to provide an explicit specification of the optimal solution or prices. Alternately, researchers have employed numerical solution techniques to solve instances of the mixed bundling problem. I reduce the multi-product pricing problem to a univariate nonlinear optimization problem in the bundle price with a global maximum that can be identified uniquely. I derive tight lower and upper bound functions for the optimal bundle price. The average of these two functions can be used as an alternate, and extremely simple, analytical expression for the optimal bundle price. Another alternative formula for the optimal bundle price is obtained by applying statistical estimation over solutions for a large space of problem instances. Both approximations are highly accurate, producing a pro t that is within half percent of the exact optimal pro t. The optimal bundle price has an elegant analytical expression (a linear function of the marginal costs of the two goods) and prices for component goods are an algebraic function of the bundle price. With a bivariate distribution in the unit square and constant unit marginal costs w1 and w2, the optimal bundle price can be written as 0:86032 0:55929(w1 w2)/2.

Generalization Bounds for Certain Class of Ranking Algorithm

The quality of ranking determines the success or failure of information retrieval and the goal of ranking is to learn a real-valued ranking function that induces a ranking or ordering over an instance space. We focus on a ranking setting which uses truth function to label each pair of instances and the ranking preferences are given randomly from some distributions on the set of possible undirected edge sets of a graph. The contribution of this paper is the given generalization bounds for such ranking algorithm via strong and weak stability. Such stabilities have lower demand than uniform stability and fit for more real applications.

MILIS: Multiple Instance Learning with Instance Selection

Multiple instance learning (MIL) is a paradigm in supervised learning that deals with the classification of collections of instances called bags. Each bag contains a number of instances from which features are extracted. The complexity of MIL is largely dependent on the number of instances in the training data set. Since we are usually confronted with a large instance space even for moderately sized real-world data sets applications, it is important to design efficient instance selection techniques to speed up the training process without compromising the performance. In this paper, we address the issue of instance selection in MIL. We propose MILIS, a novel MIL algorithm based on adaptive instance selection. We do this in an alternating optimization framework by intertwining the steps of instance selection and classifier learning in an iterative manner which is guaranteed to converge. Initial instance selection is achieved by a simple yet effective kernel density estimator on the negative instances. Experimental results demonstrate the utility and efficiency of the proposed approach as compared to the state of the art.

Propositionalized attribute taxonomies from data for data-driven construction of concise classifiers

In this paper, we consider the problem of generating concise but accurate naive Bayes classifiers using taxonomy of propositionalized attributes. For the problem, we introduce propositionalized attribute taxonomy guided naive Bayes Learner (PAT-NBL), a machine learning algorithm that effectively utilizes taxonomy to generate compact classifiers. We extend classical naive Bayes learner to the PAT-NBL algorithm that traverses over a propositionalized taxonomy to search for a locally optimal cut. PAT-NBL uses bottom-up search to find the locally optimal cut on a given taxonomy. For the evaluation of candidate cuts, we apply conditional log-likelihood, conditional minimum description length, and conditional Akaike information criterion. The detected cut enables PAT-NBL to construct an instance space which corresponds to the taxonomy and the data. That is, after PAT-NBL determines a cut according to its information-theoretic criteria, the algorithm generates a concise naive Bayes classifier based on the cut. Our experimental results on UCI Machine Learning benchmark data sets indicate that the proposed algorithm can generate naive Bayes classifiers that are more compact and often comparably accurate to those produced by standard naive Bayes learners.

Query-Level Stability of Ranking SVM for Replacement Case

The quality of ranking determines the success or failure of information retrieval and the goal of ranking is to learn a real-valued ranking function that induces a ranking or ordering over an instance space. We focus on stability and generalization ability of ranking SVM for replacement case. The query-level stability of ranking SVM for replacement case and the generalization bounds for such ranking algorithm via query-level stability by changing one element in sample set are given.

Query-Level Stability of IRSVM for Replacement Case

The quality of ranking determines the success or failure of information retrieval and the goal of ranking is to learn a real-valued ranking function that induces a ranking or ordering over an instance space. We focus on stability and generalization ability of IRSVM for replacement case. The query-level stability of IRSVM for replacement case and the generalization bounds for such ranking algorithm via query-level stability by changing one element in sample set are given.

Ancestral Recombinations Graph: A Reconstructability Perspective Using Random-Graphs Framework

We present a random graphs framework to study pedigree history in an ideal (Wright Fisher) population. This framework correlates the underlying mathematical objects in, for example, pedigree graph, mtDNA or NRY Chr tree, ARG (Ancestral Recombinations Graph), and HUD used in literature, into a single unified random graph framework. It also gives a natural definition, based solely on the topology, of an ARG, one of the most interesting as well as useful mathematical objects in this area. The random graphs framework gives an alternative parametrization of the ARG that does not use the recombination rate q and instead uses a parameter M based on the (estimate of ) the number of non-mixing segments in the extant units. This seems more natural in a setting that attempts to tease apart the population dynamics from the biology of the units. This framework also gives a purely topological definition of GMRCA, analogous to MRCA on trees (which has a purely topological description i.e., it is a root, graph-theoretically speaking, of a tree). Secondly, with a natural extension of the ideas from random-graphs we present a sampling (simulation) algorithm to construct random instances of ARG/unilinear transmission graph. This is the first (to the best of the author's knowledge) algorithm that guarantees uniform sampling of the space of ARG instances, reflecting the ideal population model. Finally, using a measure of reconstructability of the past historical events given a collection of extant sequences, we conclude for a given set of extant sequences, the joint history of local segments along a chromosome is reconstructible.

Metrics in scheduling problems

An important problem of mathematical scheduling theory is constructing metrics which can be used for designing exact and approximate solution algorithms. The introduction of metric spaces in solving NPhard problems of scheduling theory makes it possible to apply general mathematical approaches to construct approximate solutions with a guaranteed absolute error. The application of metrics in solution spaces (spaces of permutation schedules) to some problems of scheduling theory was considered in, e.g., (1, 2). Earlier, for NPhard problems with the minimiza� tion criterion {P, R, Q}|prec, r j |{L max , C max } for the maximal lateness, a metric on the instance space (of the initial parameters of the problem) was con� structed; on the basis of this metric, a general scheme for finding approximate solutions was developed (3). In this paper, we propose an approach to constructing metrics on the instance space for problems with the sum criteria , Tj, Cj, and Uj. Consider a problem of scheduling theory with

Polynomial regression under arbitrary product distributions

In recent work, Kalai, Klivans, Mansour, and Servedio (2005) studied a variant of the (Fourier) for learning under the uniform probability distribution on {0,1} n . They showed that the L 1 polynomial regression algorithm yields agnostic (tolerant to arbitrary noise) learning algorithms with respect to the class of threshold functions--under certain restricted instance distributions, including uniform on {0,1} n and Gaussian on ? n . In this work we show how all learning results based on the Low-Degree Algorithm can be generalized to give almost identical agnostic guarantees under arbitrary product distributions on instance spaces X 1×???×X n . We also extend these results to learning under mixtures of product distributions. The main technical innovation is the use of (Hoeffding) orthogonal decomposition and the extension of the noise sensitivity method to arbitrary product spaces. In particular, we give a very simple proof that threshold functions over arbitrary product spaces have ?-noise sensitivity $O(\sqrt{\delta})$ , resolving an open problem suggested by Peres (2004).

Consuming Oppression

In the U.S. today, women's bodies, more often than men's, are used as symbols of sex. Women's bodies are used to sexualize a space (for instance, go-go dancers at a night club), make a product seem sexy (e.g., in many ad campaigns for hard liquor), or simply incite sexual desire (as in pornography). It should be no surprise, then, that sex is used to sell food and that, when it is, it typically involves both sexualizing and gendering the food product. We see this in at least three ways.