Learning from imprecise and fuzzy observations: Data disambiguation through generalized loss minimization

Methods for analyzing or learning from “fuzzy data” have attracted increasing attention in recent years. In many cases, however, existing methods (for precise, non-fuzzy data) are extended to the fuzzy case in an ad-hoc manner, and without carefully considering the interpretation of a fuzzy set when being used for modeling data. Distinguishing between an ontic and an epistemic interpretation of fuzzy set-valued data, and focusing on the latter, we argue that a “fuzzification” of learning algorithms based on an application of the generic extension principle is not appropriate. In fact, the extension principle fails to properly exploit the inductive bias underlying statistical and machine learning methods, although this bias, at least in principle, offers a means for “disambiguating” the fuzzy data. Alternatively, we therefore propose a method which is based on the generalization of loss functions in empirical risk minimization, and which performs model identification and data disambiguation simultaneously. Elaborating on the fuzzification of specific types of losses, we establish connections to well-known loss functions in regression and classification. We compare our approach with related methods and illustrate its use in logistic regression for binary classification.     

交叉熵——计算语言学消歧的一种工具

歧义问题的描述和消除问题是制约计算语言学发展的瓶颈问题.将交叉熵引入计算语言学消岐领域.采用语句的真实语义作为交叉熵的训练集的先验信息,将机器翻译的语义作为测试集后验信息,计算两者的交叉熵,并以交叉熵指导对歧义的辨识和消除.实例表明,该方法简洁有效,易于计算机自适应实现,交叉熵不失为计算语言学消岐的一种较为有效的工具.     

Contextualism and conceptual disambiguation

I distinguish between Old Contextualism, New Contextualism, and the Multiple Concepts Theory. I argue that Old Contextualism cannot handle the following three problems: (i) the disquotational paradox, (ii) upward pressure resistance, (iii) inability to avoid the acceptance of skeptical conclusions. New Contextualism, in contrast, can avoid these problems. However, since New Contextualism appears to be a semanticized mirror image of MCT, it remains unclear whether it is in fact a genuine version of contextualism.     

A Network Analysis Model for Disambiguation of Names in Lists

In research and application, social networks are increasingly extracted from relationships inferred by name collocations in text-based documents. Despite the fact that names represent real entities, names are not unique identifiers and it is often unclear when two name observations correspond to the same underlying entity. One confounder stems from ambiguity, in which the same name correctly references multiple entities. Prior name disambiguation methods measured similarity between two names as a function of their respective documents. In this paper, we propose an alternative similarity metric based on the probability of walking from one ambiguous name to another in a random walk of the social network constructed from all documents. We experimentally validate our model on actor-actor relationships derived from the Internet Movie Database. Using a global similarity threshold, we demonstrate random walks achieve a significant increase in disambiguation capability in comparison to prior models. Bradley A. Malin is a Ph.D. candidate in the School of Computer Science at Carnegie Mellon University. He is an NSF IGERT fellow in the Center for Computational Analysis of Social and Organizational Systems (CASOS) and a researcher at the Laboratory for International Data Privacy. His research is interdisciplinary and combines aspects of bioinformatics, data forensics, data privacy and security, entity resolution, and public policy. He has developed learning algorithms for surveillance in distributed systems and designed formal models for the evaluation and the improvement of privacy enhancing technologies in real world environments, including healthcare and the Internet. His research on privacy in genomic databases has received several awards from the American Medical Informatics Association and has been cited in congressional briefings on health data privacy. He currently serves as managing editor of the Journal of Privacy Technology. Edoardo M. Airoldi is a Ph.D. student in the School of Computer Science at Carnegie Mellon University. Currently, he is a researcher in the CASOS group and at the Center for Automated Learning and Discovery. His methodology is based on probability theory, approximation theorems, discrete mathematics and their geometries. His research interests include data mining and machine learning techniques for temporal and relational data, data linkage and data privacy, with important applications to dynamic networks, biological sequences and large collections of texts. His research on dynamic network tomography is the state-of-the-art for recovering information about who is communicating to whom in a network, and was awarded honors from the ACM SIG-KDD community. Several companies focusing on information extraction have adopted his methodology for text analysis. He is currently investigating practical and theoretical aspects of hierarchical mixture models for temporal and relational data, and an abstract theory of data linkage. Kathleen M. Carley is a Professor of Computer Science in ISRI, School of Computer Science at Carnegie Mellon University. She received her Ph.D. from Harvard in Sociology. Her research combines cognitive science, social and dynamic networks, and computer science (particularly artificial intelligence and machine learning techniques) to address complex social and organizational problems. Her specific research areas are computational social and organization science, social adaptation and evolution, social and dynamic network analysis, and computational text analysis. Her models meld multi-agent technology with network dynamics and empirical data. Three of the large-scale tools she and the CASOS group have developed are: BioWar a city, scale model of weaponized biological attacks and response; Construct a models of the co-evolution of social and knowledge networks; and ORA a statistical toolkit for dynamic social Network data.     

MIMO声呐的双尺度旋转不变子空间波达方向估计*

为了提高单基地多输入多输出(Multiple-Input Multiple-Output, MIMO)声呐阵列的波达方向(Direction of arrival, DOA)估计性能,提出了双尺度旋转不变子空间(Dual-Resolution Estimation of Signal Parameters via Rotational Invariance Techniques, DR-ESPRIT)算法。结合MIMO阵列虚拟阵列的结构特征,首先利用ESPRIT算法通过各条虚拟线阵内、基线间距不大于半波长的子阵间的旋转不变关系得到无模糊的粗估计结果,之后利用虚拟线阵间、基线较长的子阵间的旋转不变关系得到一组有模糊的精估计结果。参考粗估计结果对精估计结果进行解模糊,最终得到高精度无模糊的角度估计结果。为了降低运算复杂度,利用该思路对降维ESPRIT算法也进行改进,提出了双尺度降维ESPRIT算法。仿真试验首先验证了与传统算法相比,双尺度类DOA估计算法能够有效提高角度估计精度。此外,还分析了MIMO声呐阵列的发射、接收阵元的幅相扰动误差对算法角度估计性能的影响。