The polaris internal representation

The Polaris Program Manipulation System is a production quality tool for source-to-source transformations and complex analysis of Fortran code. In this paper, we describe the motivations for and the implementation of Polaris' internal representation (IR). The IR is composed of a basic abstract syntax tree on top of which exist many layers of functionality. This functionality allows complex operations on the data structure. Further, the IR is designed to enforce the consistency of the internal structure in terms of both the correctness of the data structures and the correctness of the Fortran code being manipulated. In addition, operations on the IR result in the automatic updating of affected data structures such as flow information. We describe low the system's philosophies developed from its predecessor, the Delta prototyping system, and how they were implemented in Polaris' IR. We also provide a number of examples of using the Polaris system. The research described was supported by Army contract DABT63-92-C-0033. This work is not necessarily representative of the positions or policies of the Army or the Government.     

Learning Bayesian network classifiers from label proportions

This paper deals with a classification problem known as learning from label proportions. The provided dataset is composed of unlabeled instances and is divided into disjoint groups. General class information is given within the groups: the proportion of instances of the group that belong to each class.We have developed a method based on the Structural EM strategy that learns Bayesian network classifiers to deal with the exposed problem. Four versions of our proposal are evaluated on synthetic data, and compared with state-of-the-art approaches on real datasets from public repositories. The results obtained show a competitive behavior for the proposed algorithm.     

Negatively correlated neural networks for classification

This paper presents a new algorithm for designing neural network ensembles for classification problems with noise. The idea behind this new algorithm is to encourage different individual networks in an ensemble to learn different parts or aspects of the training data so that the whole ensemble can learn the whole training data better. Negatively correlated neural networks are trained with a novel correlation penalty term in the error function to encourage such specialization. In our algorithm, individual networks are trained simultaneously rather than independently or sequentially. This provides an opportunity for different networks to interact with each other and to specialize. Experiments on two real-world problems demonstrate that the new algorithm can produce neural network ensembles with good generalization ability. This work was presented, in part, at the Third International Symposium on Artificial Life and Robotics, Oita, Japan January 19–21, 1998     

Improving the generalization performance of RBF neural networks using a linear regression technique

In this paper we present a method for improving the generalization performance of a radial basis function (RBF) neural network. The method uses a statistical linear regression technique which is based on the orthogonal least squares (OLS) algorithm. We first discuss a modified way to determine the center and width of the hidden layer neurons. Then, substituting a QR algorithm for the traditional Gram–Schmidt algorithm, we find the connected weight of the hidden layer neurons. Cross-validation is utilized to determine the stop training criterion. The generalization performance of the network is further improved using a bootstrap technique. Finally, the solution method is used to solve a simulation and a real problem. The results demonstrate the improved generalization performance of our algorithm over the existing methods.     

On the implicit acquisition of a context-free grammar by a simple recurrent neural network

The performance of a simple recurrent neural network on the implicit acquisition of a context-free grammar is re-examined and found to be significantly higher than previously reported by Elman. This result is obtained although the previous work employed a multilayer extension of the basic form of simple recurrent network and restricted the complexity of training and test corpora. The high performance is traced to a well-organized internal representation of the grammatical elements, as probed by a principal-component analysis of the hidden-layer activities. From the next-symbol-prediction performance on sentences not present in the training corpus, a capacity of generalization is demonstrated.     

一种改进的选择神经网络集成方法

提出一种改进的选择神经网络集成方法,首先构造一批单个神经网络个体,分别利用Bootstrap算法产生若干个训练集并行进行训练;然后采用聚类算法计算训练好的个体网络之间的差异度和个体网络在验证集的预测精度;最后根据个体精度和个体差异度选择合适的个体网络加入集成.实验结果验证,该集成方法能较好地提高集成的预测精度和泛化能力.     

神经网络集成的设计与应用

传统的神经网络一般采用个体网络,其应用效果很大程度上取决于使用者的经验,且网络的泛化能力不强.一种改进的神经网络集成方法,为传统神经网络存在的问题提供了一个简易的解决方案.由理论分析和实验结果可以得出结论,神经网络集成方法比传统的个体网络方法的效果更好.     

Toward data representation with spiking neurons

Notable advances in the understanding of neural processing were made when sensory systems were investigated from the viewpoint of adaptation to the statistical structure of their input space. For this purpose, mathematical methods for data representation were used. Here, we point out that emphasis on the input structure has been at the cost of the biological plausibility of the corresponding neuron models which process the natural stimuli. The signal transformation of the data representation methods does not correspond well to the signal transformations happening at the single-cell level in neural systems. Hence, we now propose data representation by means of spiking neuron models. We formulate the data representation problem as an optimization problem and derive the fundamental quantities for an iterative learning scheme. This work was presented in part at the 12th International Symposium on Artificial Life and Robotics, Oita, Japan, January 25–27, 2007     

Constraint satisfaction with an object-oriented knowledge representation language

This article gives a detailed presentation of constraint satisfaction in the hybrid LAURE language. LAURE is an object-oriented language for Artificial Intelligence (AI) applications that allows the user to combine rules, constraints, and methods that cooperate on the same objects in the same program. We illustrate why this extensibility is necessary to solve some large and difficult problems by presenting a real-life application of LAURE. We describe the syntax and the various modes in which constraints may be used, as well as the tools that are proposed by LAURE to extend constraint resolution. The resolution strategy as well as some implementation details are given to explain how we obtain good performances.     

引入知识表示的图卷积网络谣言检测方法

在谣言检测的问题上,现有的研究方法无法有效地表达谣言在社交网络传播的异构图结构特征,并且没有引入外部知识作为内容核实的手段。因此,提出了引入知识表示的图卷积网络谣言检测方法,其中知识图谱作为额外先验知识来帮助核实内容真实性。采用预训练好的词嵌入模型和知识图谱嵌入模型获取文本表示后,融合图卷积网络的同时,能够在谣言传播的拓扑图中更好地进行特征提取以提升谣言检测的精确率。实验结果表明,该模型能够更好地对社交网络中的谣言进行检测。与基准模型的对比中,在Weibo数据集上的精确率达到96.1%,在Twitter15和Twitter16数据集上的F1值分别提升了3.1%和3.3%。消融实验也表明了该方法对谣言检测皆有明显提升效果,同时验证了模型的有效性和先进性。     

基于多视图知识表示和神经网络的旅游领域实体对齐方法

针对目前旅游领域实体对齐任务中的长尾实体过多和现有知识以及标注数据稀缺的问题,提出一种基于多视图知识表示和神经网络相结合的实体对齐方法。采用预训练模型完成多视图的知识表示学习,获得了实体的结构嵌入、关系嵌入和描述信息嵌入,然后利用卷积神经网络对结合了三种视图嵌入的实体综合嵌入进行相似度计算。实验精准率达到91.4%、召回率达到87.9%、综合指标F₁值达到89.6%。结果表明,该方法有效地完成了旅游领域的实体对齐任务。     

Mathematical theory of neural learning

A mathematical theory of learning is presented in a unified manner to be applicable to various architectures of networks. The theory is based on parameter modification driven by a time series of input signals generated from a stochastic information source. A network modifies its behavior such that it adapts to the environmental information structure. The theory is self-organization of a neural system. A typical discrete structure is automatically formed through continuous parameter modification by self-organization.     

Symbolic representation of user-defined time granularities

In the recent literature on time representation, an effort has been made to characterize the notion of time granularity and the relationships between granularities. The main goals are having a common framework for their specification, and allowing the interoperability of systems adopting different time granularities. This paper considers the mathematical characterization of finite and periodic time granularities, and investigates the requirements for a user-friendly symbolic formalism that could be used for their specification. Instead of proposing yet another formalism, the paper analyzes the expressiveness of known symbolic formalisms for the representation of granularities, using the mathematical characterization as a reference model. Based on this analysis, a significant extension to the collection formalism defined in [15] is proposed, in order to capture a practically interesting class of periodic granularities. This revised version was published online in June 2006 with corrections to the Cover Date.     

基于混沌变量的前向神经网络结构优化设计

提出一种关于多层前向神经网络结构的混沌优化设计方法。将混沌变量引入神经网络结构的优化搜索中，使得神经网络的隐层节点数以及所有权参数都处于混沌状态中，整个网络结构呈现为动态变化。从动态的神经网络结构中，根据性能指标来寻找一个全局最优或近似于全局最优的网络结构。仿真实验表明，采用该方案得到的神经网络结构模型对异或问题、非线性函数具有较高的逼近精度和较好的泛化能力。     

基于RBR和CBR规划中的知识表示方法研究

"知识"是"规划"的前提和基础,通过归纳对已有"知识"的表示,就成为了"规划"的先行条件.基于"规则"和"案例"的规划是各种现代规划器常用的两种规划方式,通过对现有的"规则"和"案例"的各种知识表示方法的研究,描述了其各自的优缺点,并给出了如何选择合适的知识表示方法以处理特定规划问题的方法和思想,从而更快,更好的构建能够解决实际问题的规划系统.     

Declarative representation of strategic control knowledge

Strategic (control) knowledge typically specifies how a target task is solved. Representing such knowledge declaratively remains a difficult and practical knowledge engineering challenge. The key to addressing this challenge rests on two observations. One, strategic knowledge comprises two finer types of knowledge: subgoaling knowledge used to construct the goal structure for each problem situation pertaining to a target task, and goal-sequencing knowledge used to choose which subgoal in this goal structure is to be pursued at any given moment. Second, when subgoaling knowledge is explicit and expressed in declarative ontological terms, it is possible to fully express goal-sequencing knowledge in the same declarative terms. Building on these observations, we achieve three things. First, we analyse several conventional knowledge-based applications whose subgoaling and goal-sequencing knowledge is implicit, showing that making their subgoaling knowledge explicit permits (re)representing their goal-sequencing knowledge declaratively. Among the applications analysed are MORE and NEOMYCIN. Second, upon studying the roles of goal-sequencing knowledge vis-á-vis subgoaling knowledge, we develop a declarative formalism for representing goal-sequencing knowledge. Finally, we discuss and illustrate key benefits from using our declarative formalism, including an enhanced ability to validate and reuse goal-sequencing knowledge.     

基于本体的产品设计知识表示方法

针对产品设计知识的多样性、动态化和相关性等特点,提出了一种基于本体的产品设计知识表示方法。建立了以客体、概念集、属性集、命题集和函数集为核心的知识单元,并设计了五者之间的联系,在此基础上引入了输入输出模块,以增强产品设计知识表示的全面性和灵活性。最后以圆柱形螺旋弹簧设计为例,验证了所提方法的有效性。     

S-admissible search of hierarchical classifiers with continuous feature distributions

A remedy has been found for hierarchical classifiers which relieves the tendency toward misclassification and/or ‘reject’ decisions with the Kulkarni-Kanal S-admissible search strategy, when empty bins are present in the histograms derived by discretization of feature ranges.     

Detection and differentiation of breast cancer using neural classifiers with first warning thermal sensors

Breast cancer occurs when cells in the breast begin to grow out of control and invade nearby tissues or spread throughout the body. It is one of the leading causes of death in women. Cancer development appears to generate an increase in the temperature on the breast surface. The limitations of mammography as a screening modality, especially in young women with dense breasts, necessitated the development of novel and more effective screening strategies with high sensitivity and specificity. The aim of this study was to evaluate the feasibility of discrete thermal data (DTD) as a potential tool for the early detection of the breast cancer.Our protocol uses 1170, 16-sensor data collected from 54 individuals consisting of three different kinds of breast conditions: namely, normal, benign and cancerous breast. We compared two different kinds of neural network classifiers: the feedforward neural network and the radial basis function classifier. Temperature data from the 16 temperature sensors on the surface of the two breasts (eight sensors on each side) are fed as input to the classifiers. We demonstrated a sensitivity of 84% and 91% for these classifiers (feedforward and radial basis function, respectively) with a specificity of 100%. Our classifying systems are ready to run on large data sets.     

Extraction and representation of facts from legal briefs

In knowledge-based consultation systems, the quality of the advice rendered depends on the techniques employed to represent the domain knowledge, the explanation generating capabilities, and the control strategies utilized during the consultative advice stage. The ability to understand the problem is more crucial in providing effective consultation. In this work, the emphasis is on understanding and the consequent formulation of a plausible internal representation of legal briefs. The system developed, SIFTER, reads the given input text from a legal practitioner's point of view and retrieves from it those facts that are relevant to the particular type of case on hand. In other words, it uses the domain specific knowledge to identify the type of case and to yank out the necessary information pertaining to the case. The SIFTER generates a noun-phrase processed form of the input which contains pseudo names for the proper-nouns, dates and time-intervals. The verbs in the processed input are used to check whether the case specific events have occurred or not and then the appropriate fact-containing noun-phrases are used to instantiate the relevant legal variables and, hence, to construct an internal representation of the given problem which can then be readily used by the consultative advice stage of a problem solver or analyzer. The implementation has been done in LISP culling the required domain knowledge from the Industrial Dispute Act of India.