Search

Group lasso is a regularization widely used for feature group selection with sparsity at a group level in machine learning. Training a model with the group lasso regularization, however, leads to the selection of all the groups together that are closely related to each other although their features are useful to predict a target. In this study, we propose a new regularization, group-exclusive group lasso, for automatic exclusive feature group selection. The proposed regularization aims to enforce exclusive sparsity at an inter-group level, discouraging the coincident selection of the feature groups that are group-level correlated and share predictive powers toward the targets. The proposed method aims at higher group sparsity for selecting salient feature groups only, and is applied to neural networks. We evaluate the proposed regularization in neural networks on synthetic datasets and a real-life case for virtual metrology with automatic sensor selection in semiconductor manufacturing. [ABSTRACT FROM AUTHOR]

In semiconductor manufacturing, virtual metrology (VM) is a method of predicting physical measurements of wafer qualities using in-process information from sensors on production equipment. The relevance vector machine (RVM) is a sparse Bayesian kernel machine that has been widely used for VM modeling in semiconductor manufacturing. Missing values from equipment sensors, however, preclude training an RVM model due to missing kernels from incomplete instances. Moreover, imputation for such kernels can lead to a loss of model sparsity. In this work, we propose a restricted RVM (RRVM) that selects its basis functions from only complete instances to handle incomplete data for VM. We conduct the experiments using toy data and real-life data from an etching process for wafer fabrication. The results indicate the model’s competitive prediction accuracy with massive missing data while maintaining model sparsity. Note to Practitioners—In recent decades, virtual metrology (VM) has focused on wafer fabrication in semiconductor manufacturing due to its advantages for process monitoring and automation. Typically, signals from production process equipment can predict wafer qualities in VM, which often leads to high data dimensionality. The relevance vector machine (RVM) is an algorithm that can provide a sparse solution to a Bayesian kernel method for a prediction model. Missing components in incomplete data due to sensor failures in wafer fabrication processes, however, hinder model training, and the existing approaches to handling missing data using imputation may lead to a loss of model sparsity. This article proposes a new method for RVM with incomplete data to train a model built on fully available instances by incorporating the available components of incomplete instances into model training. Using the proposed method, one can predict wafer qualities building a model trained to maintain its sparsity. Experiments indicate that the proposed model achieves competitive prediction performance and maintains model sparsity when incomplete instances are used. [ABSTRACT FROM AUTHOR]

Identifying competitors is an essential first step in creating a competitive advantage to generate superior business performance. Diverse methods for competitor identification have been introduced in the literature, but financial transactions are overlooked in these models even though such transactions among firms, as the basic economic unit of interfirm business activities, contain critical information. This article proposes a novel method for the identification of firm-level competitors using a network constructed from financial transactions among firms. To identify competitors by the intensity of interfirm competition, the proposed method is based on two dimensions of pairs of firms, resource similarity and market commonality, which are represented in a data-driven approach using a transaction network. We propose network-theoretic similarity measures to quantify the resource similarity and market commonality between firms by considering both direct and indirect transactions in a transaction network. Then, based on the two dimensions, three types of competitors are identified with respect to a focal firm: direct, potential, and indirect competitors. The proposed method is applied to toy network data and a real-world case that includes the financial transactions of firms in Korea in 2014. [ABSTRACT FROM AUTHOR]

We introduce integrated multi-stage importance assessment (IMSIA), a method for the assessment of stage importance in multi-stage processes (MSPs) characterized by a hierarchy of relationships between stages. The importance measured by IMSIA may be used as a basis to strategize intervention actions on key stages directed at influencing the process outcome. The method deconstructs the overall MSP into its local building blocks, each composed by a specific individual stage and all of its direct inputs. Within each building block, a random forest regression model and a novel conditional permutation metric are then employed to quantify the local contribution of every input to the stage in question. Global contributions are finally measured by integrating the local contribution assessments, based on the overall structure of indirect and direct relationships in the process. After a theoretical characterization of the procedure, we illustrate IMSIA on two simulated MSP, as well as on a real-world multi-stage process of industrial semiconductor manufacturing. [ABSTRACT FROM AUTHOR]

Effectively ranking patents in outlierness in a patent citation network is a crucial task for patent analysis, including as it relates to technological opportunity discovery (TOD). Previous studies in the area of TOD focus on patent textual data. In this paper, we introduce a new approach that addresses TOD via patent outlierness, leveraging both patent attributes and citations. We propose the following characteristics for patent outliers: 1) not highly clustered with other patents; 2) low node centrality within the citation network; and 3) low similarity to other patents in the network. Existing outlier ranking approaches have the drawback of not leveraging the unique characteristics of attributed patent citation networks. We propose new outlier ranking methods developed specifically for patents in attributed patent citation networks. Attribute data independently describe a patent, while citation network data relate patents to each other, thus capturing patent outlierness from two different aspects. The contributions of this paper are, given an attributed patent citation network: 1) patent clustering algorithm, and 2) method for scoring and ranking patents in outlierness. Developed methods are validated using artificial datasets. Proposed outlier ranking methods are evaluated using U.S. patents in the area of digital information and security. [ABSTRACT FROM PUBLISHER]