Automatic Speech Recognition (ASR) is the process of mapping an acoustic speech signal into a human readable text format. Traditional systems exploit the Acoustic Component of ASR using the Gaussian Mixture Model- Hidden Markov Model (GMM-HMM) approach.Deep NeuralNetwork (DNN) opens up new possibilities to overcome the shortcomings of conventional statistical algorithms. Recent studies modeled the acoustic component of ASR system using DNN in the so called hybrid DNN-HMM approach. In the context of activation functions used to model the non-linearity in DNN, Rectified Linear Units (ReLU) and maxout units are mostly used in ASR systems. This paper concentrates on the acoustic component of a hybrid DNN-HMM system by proposing an efficient activation function for the DNN network. Inspired by previous works, euclidean norm activation function is proposed to model the non-linearity of the DNN network. Such non-linearity is shown to belong to the family of Piecewise Linear (PWL) functions having distinct features. These functions can capture deep hierarchical features of the pattern. The relevance of the proposal is examined in depth both theoretically and experimentally. The performance of the developed ASR system is evaluated in terms of Phone Error Rate (PER) using TIMIT database. Experimental results achieve a relative increase in performance by using the proposed function over conventional activation functions.
Sentiment lexicons (SL) (aka lexical resources) are the repositories of one or several dictionaries that consist of known and precompiled sentiment terms. These lexicons play an important role in performing several different opinion mining tasks. The efficacy of the lexicon-based approaches in performing opinion mining (OM) tasks solely depends on selecting an appropriate opinion lexicon to analyze the text. Therefore, one has to explore the available sentiment lexicons and then select the most suitable resource. Among available resources, SentiWordNet (SWN) is the most widely used lexicon to perform tasks related to opinion mining. In SWN, each synset of WordNet is being assigned the three sentiment numerical scores; positive, negative and objective that are calculated using by a set of classifiers. In this paper, a detailed and comprehensive review of the work related to opinion mining using SentiWordNet is provided in a very distinctive way. This survey will be useful for the researchers contributing to the field of opinion mining. Following features make our contribution worthwhile and unique among the reviews of similar kind: (i) our review classifies the existing literature with respect to opinion mining tasks and subtasks (ii) it covers a very different outlook of the opinion mining field by providing in-depth discussions of the existing works at different granularity levels (word, sentences, document, aspect, clause, and concept levels) (iii) this state-ofart review covers each article in the following dimensions: the designated task performed, granularity level of the task completed, results obtained, and feature dimensions, and (iv) lastly it concludes the summary of the related articles according to the granularity levels, publishing years, related tasks (or subtasks), and types of classifiers used. In the end, major challenges and tasks related to lexicon-based approaches towards opinion mining are also discussed.
Skyline queries are extensively incorporated in various real-life applications by filtering uninteresting data objects. Sometimes, a skyline query may return so many results because it cannot control the retrieval conditions especially for highdimensional datasets. As an extension of skyline query, the kdominant skyline query reduces the control of the dimension by controlling the value of the parameter kto achieve the purpose of reducing the retrieval objects. In addition, with the continuous promotion of Bigdata applications, the data we acquired may not have the entire content that people wanted for some practically reasons of delivery failure, no power of battery, accidental loss, so that the data might be incomplete with missing values in some attributes. Obviously, the k-dominant skyline query algorithms of incomplete data depend on the user definition in some degree and the results cannot be shared. Meanwhile, the existing algorithms are unsuitable for directly used to the incomplete big data. Based on the above situations, this paper mainly studies k-dominant skyline query problem over incomplete dataset and combines this problem with the distributed structure like MapReduce environment. First, we propose an index structure over incomplete data, named incomplete data index based on dominate hierarchical tree (ID-DHT). Applying the bucket strategy, the incomplete data is divided into different buckets according to the dimensions of missing attributes. Second, we also put forward query algorithm for incomplete data in MapReduce environment, named MapReduce incomplete data based on dominant hierarchical tree algorithm (MR-ID-DHTA). The data in the bucket is allocated to the subspace according to the dominant condition by Map function. Reduce function controls the data according to the key value and returns the k-dominant skyline query result. The effective experiments demonstrate the validity and usability of our index structure and the algorithm.
Graph coloring has a wide range of real world applications, such as in the operations research, communication network, computational biology and compiler optimization fields. In our recent work , we propose a divide-andconquer approach for graph coloring, called VColor. Such an approach has three generic subroutines. (i) Graph partition subroutine: VColor partitions a graph G into a vertex cut partition (VP), which comprises a vertex cut component (VCC) and small non-overlapping connected components (CCs). (ii) Component coloring subroutine: VColor colors the VCC and the CCs by efficient algorithms. (iii) Color combination subroutine: VColor combines the local colors by exploiting the maximum matchings of color combination bigraphs (CCBs). VColor has revealed some major bottlenecks of efficiency in these subroutines. Therefore, in this paper, we propose VColor*, an approach which addresses these efficiency bottlenecks without using more colors both theoretically and experimentally. The technical novelties of this paper are the following. (i) We propose the augmented VP to index the crossing edges of the VCC and the CCs and propose an optimized CCB construction algorithm. (ii) For sparse CCs, we propose using a greedy coloring algorithm that is of polynomial time complexity in the worst case, while preserving the approximation ratio. (iii) We propose a distributed graph coloring algorithm. Our extensive experimental evaluation on real-world graphs confirms the efficiency of VColor*. In particular, VColor* is 20X and 50X faster than VColor and uses the same number of colors with VColor on the Pokec and PA datasets, respectively. VColor* also significantly outperforms the state-ofthe- art graph coloring methods.
Information networks provide a powerful representation of entities and the relationships between them. Information networks fusion is a technique for information fusion that jointly reasons about entities, links and relations in the presence of various sources. However, existing methods for information networks fusion tend to rely on a single task which might not get enough evidence for reasoning. In order to solve this issue, in this paper, we present a novel model called MC-INFM (information networks fusion model based on multi-task coordination). Different from traditional models, MC-INFM casts the fusion problem as a probabilistic inference problem, and collectively performs multiple tasks (including entity resolution, link prediction and relation matching) to infer the final result of fusion. First, we define the intra-features and the inter-features respectively and model them as factor graphs, which can provide abundant evidence to infer. Then, we use conditional random field (CRF) to learn the weight of each feature and infer the results of these tasks simultaneously by performing the maximum probabilistic inference. Experiments demonstrate the effectiveness of our proposed model.
Modern database systems desperate for the ability to support highly scalable transactions and efficient queries simultaneously for real-time applications. One solution is to utilize query optimization techniques on the on-line transaction processing (OLTP) systems. The materialized view is considered as a panacea to decrease query latency. However, it also involves the significant cost of maintenance which trades away transaction performance. In this paper, we examine the design space and conclude several design features for the implementation of a view on a distributed log-structured merge-tree (LSMtree), which is a well-known structure for improving data write performance. As a result, we develop two incremental view maintenance (IVM) approaches on LSM-tree. One avoids join computation in view maintenance transactions. Another with two optimizations is proposed to decouple the view maintenance with the transaction process. Under the asynchronous update, we also provide consistency queries for views. Experiments on TPC-H benchmark show our methods achieve better performance than straightforward methods on different workloads.
Richly formatted documents, such as financial disclosures, scientific articles, government regulations, widely exist on Web. However, since most of these documents are only for public reading, the styling information inside them is usually missing, making them improper or even burdensome to be displayed and edited in different formats and platforms. In this study we formulate the task of document styling restoration as an optimization problem, which aims to identify the styling settings on the document elements, e.g., lines, table cells, text, so that rendering with the output styling settings results in a document, where each element inside it holds the (closely) exact position with the one in the original document. Considering that each styling setting is a decision, this problem can be transformed as a multi-step decision-making task over all the document elements, and then be solved by reinforcement learning. Specifically, Monte-Carlo Tree Search (MCTS) is leveraged to explore the different styling settings, and the policy function is learnt under the supervision of the delayed rewards. As a case study, we restore the styling information inside tables, where structural and functional data in the documents are usually presented. Experiment shows that, our best reinforcement method successfully restores the stylings in 87.65% of the tables, with 25.75% absolute improvement over the greedymethod.We also discuss the tradeoff between the inference time and restoration success rate, and argue that although the reinforcement methods cannot be used in real-time scenarios, it is suitable for the offline tasks with high-quality requirement. Finally, this model has been applied in a PDF parser to support cross-format display.
We consider image transformation problems, and the objective is to translate images from a source domain to a target one. The problem is challenging since it is difficult to preserve the key properties of the source images, and to make the details of target being as distinguishable as possible. To solve this problem, we propose an informative coupled generative adversarial networks (ICoGAN). For each domain, an adversarial generator-and-discriminator network is constructed. Basically, we make an approximately-shared latent space assumption by a mutual information mechanism, which enables the algorithm to learn representations of both domains in unsupervised setting, and to transform the key properties of images from source to target.Moreover, to further enhance the performance, a weightsharing constraint between two subnetworks, and different level perceptual losses extracted from the intermediate layers of the networks are combined. With quantitative and visual results presented on the tasks of edge to photo transformation, face attribute transfer, and image inpainting, we demonstrate the ICo- GAN’s effectiveness, as compared with other state-of-the-art algorithms.
Survey generation aims to generate a summary from a scientific topic based on related papers. The structure of papers deeply influences the generative process of survey, especially the relationships between sentence and sentence, paragraph and paragraph. In principle, the structure of paper can influence the quality of the summary. Therefore, we employ the structure of paper to leverage contextual information among sentences in paragraphs to generate a survey for documents. In particular, we present a neural document structure model for survey generation.We take paragraphs as units, and model sentences in paragraphs, we then employ a hierarchical model to learn structure among sentences, which can be used to select important and informative sentences to generate survey. We evaluate our model on scientific document data set. The experimental results show that our model is effective, and the generated survey is informative and readable.
The World Wide Web generates more and more data with links and node contents, which are always modeled as attributed networks. The identification of network communities plays an important role for people to understand and utilize the semantic functions of the data. A few methods based on non-negative matrix factorization (NMF) have been proposed to detect community structure with semantic information in attributed networks. However, previous methods have not modeled some key factors (which affect the link generating process together), including prior information, the heterogeneity of node degree, as well as the interactions among communities. The three factors have been demonstrated to primarily affect the results. In this paper, we propose a semi-supervised community detection method on attributed networks by simultaneously considering these three factors. First, a semi-supervised non-negative matrix tri-factorization model with node popularity (i.e., PSSNMTF) is designed to detect communities on the topology of the network. And then node contents are integrated into the PSSNMTF model to find the semantic communities more accurately, namely PSSNMTFC. Parameters of the PSSNMTFC model is estimated by using the gradient descent method. Experiments on some real and artificial networks illustrate that our new method is superior over some related stateof- the-art methods in terms of accuracy.
Crowdsourcing has been a helpful mechanism to leverage human intelligence to acquire useful knowledge.However, when we aggregate the crowd knowledge based on the currently developed voting algorithms, it often results in common knowledge that may not be expected. In this paper, we consider the problem of collecting specific knowledge via crowdsourcing. With the help of using external knowledge base such as WordNet, we incorporate the semantic relations between the alternative answers into a probabilisticmodel to determine which answer is more specific. We formulate the probabilistic model considering both worker’s ability and task’s difficulty from the basic assumption, and solve it by the expectation-maximization (EM) algorithm. To increase algorithm compatibility, we also refine our method into semi-supervised one. Experimental results show that our approach is robust with hyper-parameters and achieves better improvement thanmajority voting and other algorithms when more specific answers are expected, especially for sparse data.
Technical debt is a metaphor for seeking short-term gains at expense of long-term code quality. Previous studies have shown that self-admitted technical debt, which is introduced intentionally, has strong negative impacts on software development and incurs high maintenance overheads. To help developers identify self-admitted technical debt, researchers have proposed many state-of-the-art methods. However, there is still room for improvement about the effectiveness of the current methods, as self-admitted technical debt comments have the characteristics of length variability, low proportion and style diversity. Therefore, in this paper, we propose a novel approach based on the bidirectional long short-term memory (BiLSTM) networks with the attention mechanism to automatically detect self-admitted technical debt by leveraging source code comments. In BiLSTM, we utilize a balanced cross entropy loss function to overcome the class unbalance problem. We experimentally investigate the performance of our approach on a public dataset including 62, 566 code comments from ten open source projects. Experimental results show that our approach achieves 81.75% in terms of precision, 72.24% in terms of recall and 75.86% in terms of F1-score on average and outperforms the state-of-the-art text mining-based method by 8.14%, 5.49% and 6.64%, respectively.
Emerging persistent memory technologies, like PCM and 3D XPoint, offer numerous advantages, such as higher density, larger capacity, and better energy efficiency, compared with the DRAM. However, they also have some drawbacks, e.g., slower access speed, limited write endurance, and unbalanced read/write latency. Persistent memory technologies provide both great opportunities and challenges for operating systems. As a result, a large number of solutions have been proposed. With the increasing number and complexity of problems and approaches, we believe this is the right moment to investigate and analyze these works systematically.
To this end, we perform a comprehensive and in-depth study on operating system support for persistent memory within three steps. First, we present an overview of how to build the operating system on persistent memory from three perspectives: system abstraction, crash consistency, and system reliability. Then, we classify the existing research works into three categories: storage stack, memory manager, and OS-bypassing library. For each category, we summarize the major research topics and discuss these topics deeply. Specifically, we present the challenges and opportunities in each topic, describe the contributions and limitations of proposed approaches, and compare these solutions in different dimensions. Finally, we also envision the future operating system based on this study.
Exploring the interleaving space of a multithreaded program to efficiently detect concurrency bugs is important but also difficult because of the astronomically many thread schedules. This paper presents a novel framework to decompose a thread schedule generator that explores the interleaving space into the composition of a basic generator and its extension under the “small interleaving hypothesis”. Under this framework, we in-depth analyzed research work on interleaving space exploration, illustrated how to design an effective schedule generator, and shed light on future research opportunities.