Authorship：Lead author   Publisher：Springer  

DOI： 10.1007/978-3-319-97310-4_18

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Verlag  

This paper proposes a method of scoring sequences generated by recurrent neural network (RNN) for automatic Tanka composition. Our method gives sequences a score based on topic assignments provided by latent Dirichlet allocation (LDA). When many word tokens in a sequence are assigned to the same topic, we give the sequence a high score. While a scoring of sequences can also be achieved by using RNN output probabilities, the sequences having large probabilities are likely to share much the same subsequences and thus are doomed to be deprived of diversity. The experimental results, where we scored Japanese Tanka poems generated by RNN, show that the top-ranked sequences selected by our method were likely to contain a wider variety of subsequences than those selected by RNN output probabilities.

DOI： 10.1007/978-3-319-93713-7_33

Scopus

researchmap

Publisher：ACM  

DOI： 10.1145/3224207.3224214

researchmap

Publisher：Springer  

DOI： 10.1007/978-3-030-05090-0_25

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Verlag  

This paper proposes a new method for estimating the word probabilities in latent Dirichlet allocation (LDA). LDA uses a Dirichlet distribution as the prior for the per-document topic discrete distributions. While another Dirichlet prior can be introduced for the per-topic word discrete distributions, point estimations may lead to a better evaluation result, e.g. in terms of test perplexity. This paper proposes a method for the point estimation of the per-topic word probabilities in LDA by using multilayer perceptron (MLP). Our point estimation is performed in an online manner by mini-batch gradient ascent. We compared our method to the baseline method using a perceptron with no hidden layers and also to the collapsed Gibbs sampling (CGS). The evaluation experiment showed that the test perplexity of CGS could not be improved in almost all cases. However, there certainly were situations where our method achieved a better perplexity than the baseline. We also discuss a usage of our method as word embedding.

DOI： 10.1007/978-3-319-67274-8_12

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

This paper proposes a method for exploring out-of-vocabulary (OOV) words from Myanmar text by using maximal substrings. Our main purpose is to find OOV words that can be added into the Myanmar dictionary. The outcome of our method are new compound words that do not exist in the Myanmar dictionary. Our method consists of two steps. In the first step, we extract maximal substrings, i.e., the substrings whose number of occurrences are decreased only after appending a character before or after them, from Myanmar news articles. In the second step, we make the post processing of maximal substrings, because the results obtained by maximal substrings contain noisy characters. Our post-processing is threefold. First, we reduce the number of maximal substrings. Second, we remove maximal substrings whose prefixes and suffixes are meaningless characters. Third, we find OOV words that are the substrings consisting of the two words from the existing dictionary. Consequently, we obtain the substrings as candidates of new compound words that can be inserted into the existing Myanmar dictionary after being scrutinized by native speakers. We evaluate the accuracy of new compound words by using the subjective perspective. It is found that our results do seem promising. We appeal that new compound words obtained by our method are useful for expressing the words as a single unit of meaning that can be utilized in Myanmar text effectively.

DOI： 10.1109/IIAI-AAI.2016.73

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/iiaiaai/iiaiaai2016.html#conf/iiaiaai/WinM16

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

This paper proposes a method for proper names extraction from Myanmar text by using latent Dirichlet allocation (LDA). Our method aims to extract proper names that provide important information on the contents of Myanmar text. Our method consists of two steps. In the first step, we extract topic words from Myanmar news articles by using LDA. In the second step, we make a post-processing, because the resulting topic words contain some noisy words. Our post-processing, first of all, eliminates the topic words whose prefixes are Myanmar digits and suffixes are noun and verb particles. We then remove the duplicate words and discard the topic words that are contained in the existing dictionary. Consequently, we obtain the words as candidate of proper names, namely personal names, geographical names, unique object names, organization names, single event names, and so on. The evaluation is performed both from the subjective and quantitative perspectives. From the subjective perspective, we compare the accuracy of proper names extracted by our method with those extracted by latent semantic indexing (LSI) and rule-based method. It is shown that both LSI and our method can improve the accuracy of those obtained by rule-based method. However, our method can provide more interesting proper names than LSI. From the quantitative perspective, we use the extracted proper names as additional features in K-means clustering. The experimental results show that the document clusters given by our method are better than those given by LSI and rule-based method in precision, recall and F-score.

DOI： 10.1109/TAAI.2016.7880176

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER INT PUBLISHING AG  

This paper proposes a new inference for the latent Dirichlet allocation (LDA) [4]. Our proposal is an instance of the stochastic gradient variational Bayes (SGVB) [9,13]. SGVB is a general framework for devising posterior inferences for Bayesian probabilistic models. Our aim is to show the effectiveness of SGVB by presenting an example of SGVB-type inference for LDA, the best-known Bayesian model in text mining. The inference proposed in this paper is easy to implement from scratch. A special feature of the proposed inference is that the logistic normal distribution is used to approximate the true posterior. This is counterintuitive, because we obtain the Dirichlet distribution by taking the functional derivative when we lower bound the log evidence of LDA after applying a mean field approximation. However, our experiment showed that the proposed inference gave a better predictive performance in terms of test set perplexity than the inference using the Dirichlet distribution for posterior approximation. While the logistic normal is more complicated than the Dirichlet, SGVB makes the manipulation of the expectations with respect to the posterior relatively easy. The proposed inference was better even than the collapsed Gibbs sampling [6] for not all but many settings consulted in our experiment. It must be worthwhile future work to devise a new inference based on SGVB also for other Bayesian models.

DOI： 10.1007/978-3-319-42089-9_17

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Verlag  

This paper proposes a new inference for the correlated topic model (CTM) [3]. CTM is an extension of LDA [4] for modeling correlations among latent topics. The proposed inference is an instance of the stochastic gradient variational Bayes (SGVB) [7,8]. By constructing the inference network with the diagonal logistic normal distribution, we achieve a simple inference. Especially, there is no need to invert the covariance matrix explicitly. We performed a comparison with LDA in terms of predictive perplexity. The two inferences for LDA are considered: the collapsed Gibbs sampling (CGS) [5] and the collapsed variational Bayes with a zero-order Taylor expansion approximation (CVB0) [1]. While CVB0 for LDA gave the best result, the proposed inference achieved the perplexities comparable with those of CGS for LDA.

DOI： 10.1007/978-3-319-45817-5_39

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

This paper provides a heuristic pretraining for topic models. While we consider latent Dirichlet allocation (LDA) here, our pretraining can be applied to other topic models. Basically, we use collapsed Gibbs sampling (CGS) to update the latent variables. However, after every iteration of CGS, we regard the latent variables as observable and construct another LDA over them, which we call LDA over LDA (LoL). We then perform the following two types of updates: the update of the latent variables in LoL by CGS and the update of the latent variables in LDA based on the result of the preceding update of the latent variables in LoL. We perform one iteration of CGS for LDA and the above two types of updates alternately only for a small, earlier part of the inference. That is, the proposed method is used as a pretraining. The pretraining stage is followed by the usual iterations of CGS for LDA. The evaluation experiment shows that our pretraining can improve test set perplexity.

DOI： 10.1007/978-3-319-19066-2_34

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER INT PUBLISHING AG  

This paper presents a novel topic model for traffic speed analysis in the urban environment. Our topic model is special in that the parameters for encoding the following two domain-specific aspects of traffic speeds are introduced. First, traffic speeds are measured by the sensors each having a fixed location. Therefore, it is likely that similar measurements will be given by the sensors located close to each other. Second, traffic speeds show a 24-hour periodicity. Therefore, it is likely that similar measurements will be given at the same time point on different days. We model these two aspects with Gaussian process priors and make topic probabilities location-and time-dependent. In this manner, our model utilizes the metadata of the traffic speed data. We offer a slice sampling to achieve less approximation than variational Bayesian inferences. We present an experimental result where we use the traffic speed data provided by New York City.

DOI： 10.1007/978-3-319-26135-5_10

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

This paper proposes a method for exploring technical phrase frames by extracting word n-grams that match our information needs and interests from research paper titles. Technical phrase frames, the outcome of our method, are phrases with wildcards that may be substituted for any technical term. Our method, first of all, extracts word trigrams from research paper titles and constructs a co-occurrence graph of the trigrams. Even by simply applying PageRank algorithm to the co-occurrence graph, we obtain the trigrams that can be regarded as technical keyphrases at the higher ranks in terms of PageRank score. In contrast, our method assigns weights to the edges of the co-occurrence graph based on Jaccard similarity between trigrams and then apply weighted PageRank algorithm. Consequently, we obtain widely different but more interesting results. While the top-ranked trigrams obtained by unweighted PageRank have just a self-contained meaning, those obtained by our method are technical phrase frames, i.e., a word sequence that forms a complete technical phrase only after putting a technical word (or words) before or/and after it. We claim that our method is a useful tool for discovering important phraseological patterns, which can expand query keywords for improving information retrieval performance and can also work as candidate phrasings in technical writing to make our research papers attractive.

DOI： 10.1109/WAINA.2015.37

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

This paper provides an application of sparse additive generative models (SAGE) for temporal topic analysis. In our model, called ChronoSAGE, topic modeling results are diversified chronologically by using document timestamps. That is, word tokens are generated not only in a topic-specific manner, but also in a time-specific manner. We firstly compare ChronoSAGE with latent Dirichlet allocation (LDA) in terms of pointwise mutual information to show its practical effectiveness. We secondly give an example of time-differentiated topics, obtained by ChronoSAGE as word lists, to show its usefulness in trend detection.

DOI： 10.1007/978-3-319-08010-9_51

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Verlag  

We propose a probabilistic model for traffic speed data. Our model inherits two key features from latent Dirichlet allocation (LDA). Firstly, unlike e.g. stock market data, lack of data is often perceived for traffic speed data due to unexpected failure of sensors or networks. Therefore, we regard speed data not as a time series, but as an unordered multiset in the same way as LDA regards documents not as a sequence, but as a bag of words. This also enables us to analyze co-occurrence patterns of speed data regardless of their positions along the time axis. Secondly, we regard a daily set of speed data gathered from the same sensor as a document and model it not with a single distribution, but with a mixture of distributions as in LDA. While each such distribution is called topic in LDA, we call it patch to remove text-mining connotation and name our model Patchy. This approach enables us to model speed co-occurrence patterns effectively. However, speed data are non-negative real. Therefore, we use Gamma distributions in place of multinomial distributions. Due to these two features, Patchy can reveal context dependency of traffic speed data. For example, a 60 mph observed on Sunday can be assigned to a patch different from that to which a 60 mph on Wednesday is assigned. We evaluate this context dependency through a binary classification task, where test data are classified as either weekday data or not. We use real traffic speed data provided by New York City and compare Patchy with the baseline method, where a simpler data model is applied. © 2014 Springer International Publishing Switzerland.

DOI： 10.1007/978-3-319-07467-2_8

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

This paper reports a pilot study where we link different types of data for explaining prices. In this study, we link the apartment rent data with the publicly accessible location data of landmarks like supermarkets, hospitals, etc. We apply the regression analysis to find the most important factor determining the apartment rents. We claim that the results of this type of spatial data mining can enhance the user experience in the apartment search system, because we can indicate a rationale behind pricing as additional information to users and thus can make them more confident in their choices.

DOI： 10.1109/GCCE.2014.7031088

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Successful research collaborations may facilitate major outcomes in science and their applications. Thus, identifying effective collaborators may be a key factor that affects success. However, it is very difficult to identify potential collaborators and it is particularly difficult for young researchers who have less knowledge about other researchers and experts in their research domain. This study introduces and defines the problem of collaborator recommendation for 'isolated' researchers who have no links with others in coauthor networks. Existing approaches such as link-based and content-based methods may not be suitable for isolated researchers because of their lack of links and content information. Thus, we propose a new approach that uses additional information as new features to make recommendations, i.e., the strength of the relationship between organizations, the importance rating, and the activity scores of researchers. We also propose a new method for evaluating the quality of collaborator recommendations. We performed experiments by crawling publications from the Microsoft Academic Search website. The metadata were extracted from these publications, including the year, authors, organizational affiliations of authors, citations, and references. The metadata from publications between 2001 and 2005 were used as the training data while those from 2006 to 2011 were used for validation. The experimental results demonstrated the effectiveness and efficiency of our proposed approach.

DOI： 10.1109/WAINA.2014.105

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

We propose a new probabilistic model for constructing efficient prototypes of handwritten digit images. We assume that all digit images are of the same size and obtain one color histogram for each pixel by counting the number of occurrences of each color over multiple images. For example, when we conduct the counting over the images of digit "5", we obtain a set of histograms as a prototype of digit "5". After normalizing each histogram to a probability distribution, we can classify an unknown digit image by multiplying probabilities of the colors appearing at each pixel of the unknown image. We regard this method as the baseline and compare it with a method using our probabilistic model called Multinomialized Subset Infinite Relational Model (MSIRM), which gives a prototype, where color histograms are clustered column- and row-wise. The number of clusters is adjusted flexibly with Chinese restaurant process. Further, MSIRM can detect irrelevant columns and rows. An experiment, comparing our method with the baseline and also with a method using Dirichlet process mixture, revealed that MSIRM could neatly detect irrelevant columns and rows at peripheral part of digit images. That is, MSIRM could "trim" irrelevant part. By utilizing this trimming, we could speed up classification of unknown images. © 2013 Springer Science+Business Media Dordrecht(Outside the USA).

DOI： 10.1007/978-94-007-6738-6_16

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

In this paper, we provide a revised inference for correlated topic model (CTM) [3]. CTM is proposed by Blei et al. for modeling correlations among latent topics more expressively than latent Dirichlet allocation (LDA) [2] and has been attracting attention of researchers. However, we have found that the variational inference of the original paper is unstable due to almost-singularity of the covariance matrix when the number of topics is large. This means that we may be reluctant to use CTM for analyzing a large document set, which may cover a rich diversity of topics. Therefore, we revise the inference and improve its quality. First, we modify the formula for updating the covariance matrix in a manner that enables us to recover the original inference by adjusting a parameter. Second, we regularize posterior parameters for reducing a side effect caused by the formula modification. While our method is based on a heuristic intuition, an experiment conducted on large document sets showed that it worked effectively in terms of perplexity. © 2013 Springer-Verlag Berlin Heidelberg.

DOI： 10.1007/978-3-642-39068-5-54

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

This paper proposes a new approach for handwritten digit image classification using a nonparametric Bayesian probabilistic model, called multinomialized subset infinite relational model (MSIRM). MSIRM realizes a three-way clustering, i.e., a simultaneous clustering of digit images, pixel columns, and pixel rows, where the numbers of clusters are adjusted automatically with Chinese restaurant process (CRP). We obtain MSIRM as a modification of subset infinite relational model (SIRM) by Ishiguro et al. [4] While this modification is straightforward, our application of MSIRM to handwritten digit image classification leads to an impressive result. To represent a large number of training digit images in a compact form, we cluster the training images and then classify a test image to the class of the cluster most similar to the test image. By extending this line of thought, MSIRM clusters not only digit images but also pixel columns and pixel rows to obtain a more compact representation. With this three-way clustering, we achieved 2.95% and 5.38% test error rates for MNIST and USPS datasets, respectively. © Springer-Verlag 2013.

DOI： 10.1007/978-3-642-42051-1_20

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

This paper provides experimental results showing that we can use maximal substrings as elementary building blocks of documents in place of the words extracted by a current state-of-the-art supervised word extraction. Maximal substrings are defined as the substrings each giving a smaller number of occurrences even by appending only one character to its head or tail. The main feature of maximal substrings is that they can be extracted quite efficiently in an unsupervised manner. We extract maximal substrings from a document set and represent each document as a bag of maximal substrings. We also obtain a bag of words representation by using a state-of-the-art supervised word extraction over the same document set. We then apply the same document clustering method to both representations and obtain two clustering results for a comparison of their quality. We adopt a Bayesian document clustering based on Dirichlet compound multinomials for avoiding overfitting. Our experiment shows that the clustering quality achieved with maximal substrings is acceptable enough to use them in place of the words extracted by a supervised word extraction.

DOI： 10.1007/978-3-642-29958-2_2

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：ACM  

This paper provides a topic model for extracting topic evolutions as a corpus-wide transition matrix among latent topics. Recent trends in text mining point to a high demand for exploiting metadata. Especially, exploitation of reference relationships among documents induced by hyperlinking Web pages, citing scientific articles, tumblring blog posts, retweeting tweets, etc., is put in the foreground of the effort for an effective mining. We focus on scholarly activities and propose a topic model for obtaining a corpus-wide view on how research topics evolve along citation relationships. Our model, called TERESA, extends latent Dirichlet allocation (LDA) by introducing a corpus-wide topic transition probability matrix, which models reference relationships as transitions among topics. Our approximated variational inference updates LDA posteriors and topic transition posteriors alternately. The main issue is execution time amounting to O(MK2), where K is the number of topics and M is that of links in citation network. Therefore, we accelerate the inference with Nvidia CUDA compatible GPUs. We compare the effectiveness of TERESA with that of LDA by introducing a new measure called diversity plus focusedness (D+F). We also present topic evolution examples our method gives. © 2012 ACM.

DOI： 10.1145/2396761.2398465

Scopus

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/cikm/cikm2012.html#conf/cikm/MasadaT12

Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

This paper introduces a novel approach for large-scale unsupervised segmentation of bibliographic elements. Our problem is to segment a word token sequence representing a citation into subsequences each corresponding to a different bibliographic element, e.g. authors, paper title, journal name, publication year, etc. Obviously, each bibliographic element should be represented by contiguous word tokens. We call this constraint contiguity constraint. Therefore, we should infer a sequence of assignments of word tokens to bibliographic elements so that this constraint is satisfied. Many HMM-based methods solve this problem by prescribing fixed transition patterns among bibliographic elements. In this paper, we use generalized Mallows models (GMM) in a Bayesian multi-topic model, effectively applied to document structure learning by Chen et al. [4], and infer a permutation of latent topics each of which can be interpreted as one among the bibliographic elements. According to the inferred permutation, we arrange the order of the draws from a multinomial distribution defined over topics. In this manner, we can obtain an ordered sequence of topic assignments satisfying contiguity constraint. We do not need to prescribe any transition patterns among bibliographic elements. We only need to specify the number of bibliographic elements. However, the method proposed by Chen et al. works for our problem only after introducing modification. The main contribution of this paper is to propose strategies to make their method work also for our problem. © 2011 Springer-Verlag.

DOI： 10.1007/978-3-642-24396-7_20

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

This paper provides a new approach to topical trend analysis. Our aim is to improve the generalization power of latent Dirichlet allocation (LDA) by using document timestamps. Many previous works model topical trends by making latent topic distributions time-dependent. We propose a straightforward approach by preparing a different word multinomial distribution for each time point. Since this approach increases the number of parameters, overfitting becomes a critical issue. Our contribution to this issue is two-fold. First, we propose an effective way of defining Dirichlet priors over the word multinomials. Second, we propose a special scheduling of variational Bayesian (VB) inference. Comprehensive experiments with six datasets prove that our approach can improve LDA and also Topics over Time, a well-known variant of LDA, in terms of test data perplexity in the framework of VB inference.

DOI： 10.1007/978-3-642-20841-6-36

DOI： 10.1007/978-3-642-20841-6_36

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：INSTICC-INST SYST TECHNOLOGIES INFORMATION CONTROL & COMMUNICATION  

This paper provides experimental results showing how we can use maximal substrings as elementary features in document clustering. We extract maximal substrings, i.e., the substrings each giving a smaller number of occurrences even after adding only one character at its head or tail, from the given document set and represent each document as a bag of maximal substrings after reducing the variety of maximal substrings by a simple frequency-based selection. This extraction can be done in an unsupervised manner. Our experiment aims to compare bag of maximal substrings representation with bag of words representation in document clustering. For clustering documents, we utilize Dirichlet compound multinomials, a Bayesian version of multinomial mixtures, and measure the results by F-score. Our experiment showed that maximal substrings were as effective as words extracted by a dictionary-based morphological analysis for Korean documents. For Chinese documents, maximal substrings were not so effective as words extracted by a supervised segmentation based on conditional random fields. However, one fourth of the clustering results given by bag of maximal substrings representation achieved F-scores better than the mean F-score given by bag of words representation. It can be said that the use of maximal substrings achieved an acceptable performance in document clustering.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

This paper proposes a semi-supervised bibliographic element segmentation. Our input data is a large scale set of bibliographic references each given as an unsegmented sequence of word tokens. Our problem is to segment each reference into bibliographic elements, e.g. authors, title, journal, pages, etc. We solve this problem with an LDA-like topic model by assigning each word token to a topic so that the word tokens assigned to the same topic refer to the same bibliographic element. Topic assignments should satisfy contiguity constraint, i.e., the constraint that the word tokens assigned to the same topic should be contiguous. Therefore, we proposed a topic model in our preceding work [8] based on the topic model devised by Chen et al. [3]. Our model extends LDA and realizes unsupervised topic assignments satisfying contiguity constraint. The main contribution of this paper is the proposal of a semi-supervised learning for our proposed model. We assume that at most one third of word tokens are already labeled. In addition, we assume that a few percent of the labels may be incorrect. The experiment showed that our semi-supervised learning improved the unsupervised learning by a large margin and achieved an over 90% segmentation accuracy.

DOI： 10.1007/978-3-642-24826-9_11

researchmap

Publishing type：Research paper (scientific journal)   Publisher：Association for Computing Machinery (ACM)  

DOI： 10.1145/1926367.1926375

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE COMPUTER SOC  

This paper presents infinite latent process decomposition (iLPD), a new microarray analysis method, as an extension of latent process decomposition in Our method assumes an infinite number of latent processes. Further, our new collapsed variational Bayesian inference improves the inference proposed in [2] in the treatment of Dirichlet hyperparameters. We also give the results of the comparison experiment.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

In this paper, we propose a new method for topical trend analysis. We model topical trends by per-topic Beta distributions as in Topics over Time (TOT), proposed as an extension of latent Dirichlet allocation (LDA). However, TOT is likely to overfit to timestamp data in extracting latent topics. Therefore; we apply prior distributions to Beta distributions in TOT. Since Beta distribution has no conjugate prior; we devise a trick, where we set one among the two parameters of each per-topic Beta distribution to one based on a Bernoulli trial and apply Gamma distribution as a conjugate prior. Consequently; we can marginalize out the parameters of Beta distributions and thus treat; timestamp data in a Bayesian fashion. In the evaluation experiment, we compare our method with LDA and TOT in link detection task on TDT4 dataset. We use word predictive probabilities as term weights and estimate document similarities by using those weights in a TFIDF-like scheme. The results show that our method achieves a moderate fitting to timestamp data.

DOI： 10.1007/978-3-642-13318-3_38

researchmap

Publishing type：Research paper (scientific journal)  

Recently, general-purpose computation on graphics processing units (GPGPU) has become an increasingly popular field of study as graphics processing units (GPUs) continue to be proposed as high performance and relatively low cost implementation platforms for scientific computing applications. Among these applications figure astrophysical N-bodysimulations, which form one of the most challenging problems in computational science. However, in most reported studies, a simple O(N2)algorithm was used for GPGPUs, and the resulting performances were not observed to be better than those of conventional CPUs that were based on more optimized O(N log N) algorithms such as the tree algorithm or the particle-particle particle-mesh algorithm. Because of the difficulty in getting efficient implementations of such algorithms on GPUs, a GPU cluster had no practical advantage over general-purpose PC clusters for N-bodysimulations. In this paper, we report a new method for efficient parallel implementation of the tree algorithm on GPUs. Our novel tree code allows the realization of an N-bodysimulation on a GPU cluster at a much higher performance than that on general PC clusters. We practically performed a cosmological simulation with 562 million particles on a GPU cluster using 128 NVIDIA GeForce 8800GTS GPUs at an overall cost of 168172 $. We obtained a sustained performance of 20.1 Tflops, which when normalized against a general-purpose CPU implementation leads to a performance of 8.50 Tflops. The achieved cost/performance was hence a mere $19.8 /Gflops which shows the high competitiveness of GPGPUs. © 2009 Springer-Verlag.

DOI： 10.1007/s00450-009-0089-1

Scopus

researchmap

Language：Japanese   Publisher：The Institute of Image Information and Television Engineers  

The Phase Only Correlation (POC) method demonstrates high robustness and subpixel accuracy in the pattern matching and the image registration. However, there is a disadvantage in computational speed because of the calculation of 2D-FFT etc. We have proposed a novel approach to accelerate POC method using GPU to solve the calculation cost problem. Using our GPU-based POC implementation, each POC calculation can be done within 2.36 seconds for 256×256 pixels, within 7.92 seconds for 512×512 pixels, and 27.65 seconds for 1024×1024 pixels.

DOI： 10.11485/itetr.33.6.0_201

CiNii Article

researchmap

Other Link： http://hdl.handle.net/10069/22664

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

In this paper, we propose a new probabilistic model, Bay of Timestamps (BoT), for chronological text mining. BoT is an extension of latent Dirichlet allocation (LDA), and has two remarkable features when compared with a previously proposed Topics over Time (ToT), which is also an extension of LDA. First, we can avoid overfitting to temporal data, because temporal data are modeled in a Bayesian manner similar to word frequencies. Second, BoT has a conditional probability where no functions requiring time-consuming computations appear. The experiments using newswire documents show that BoT achieves more moderate fitting to temporal data in shorter execution time than ToT.

DOI： 10.1007/978-3-642-00672-2-51

DOI： 10.1007/978-3-642-00672-2_51

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

In this paper, we propose an acceleration of collapsed variational Bayesian (CVB) inference for latent Dirichlet allocation (LDA) by using Nvidia CUDA compatible devices. While LIDA is an efficient Bayesian multi-topic document model, it requires complicated computations for parameter estimation in comparison with other simpler document models, e.g. probabilistic latent semantic indexing, etc. Therefore, we accelerate CVB inference, an efficient deterministic inference method for LIDA, with Nvidia CUDA. In the evaluation experiments, we used a set of 50,000 documents and a set of 10,000 images. We could obtain inference results comparable to sequential CVB inference.

DOI： 10.1007/978-3-642-02568-6_50

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：ACM  

This paper presents a new Bayesian topical trend analysis. We regard the parameters of topic Dirichlet priors in latent Dirichlet allocation as a function of document timestamps and optimize the parameters by a gradient-based algorithm. Since our method gives similar hyperparameters to the documents having similar timestamps, topic assignment in collapsed Gibbs sampling is affected by timestamp similarities. We compute TFIDF-based document similarities by using a result of collapsed Gibbs sampling and evaluate our proposal by link detection task of Topic Detection and Tracking. Copyright 2009 ACM.

DOI： 10.1145/1645953.1646242

Scopus

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/cikm/cikm2009.html#conf/cikm/MasadaFTHSO09

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

This paper provides a new method for multi-topic Bayesian analysis for microarray data. Our method achieves a further maximization of lower bounds in a marginalized variational Bayesian inference (MVB) for Latent Process Decomposition (LPD), which is an effective probabilistic model for microarray data. In our method, hyperparameters in LPD are updated by empirical Bayes point estimation. The experiments based on microarray data of realistically large size show efficiency of our hyperparameter reestimation technique.

DOI： 10.1007/978-3-642-03348-3_26

researchmap

Language：Japanese   Publisher：The Institute of Image Information and Television Engineers  

In this paper, we propose a GPU acceleration of multi-topic extraction from images by using LDA (latent Dirichlet allocation). LDA is originally proposed as a probabilistic model for documents by Blei et al. In recent days, LDA is applied to multimedia information other than documents. We provide the results of experiments where we apply LDA to Professor Wang's 10,000 test images and extract multiple visional topics. We adpot collapsed variational Bayesian inference method for LDA and accelerate this by using Nvidia CUDA compatible GPU devices.

DOI： 10.11485/itetr.32.54.0_1

CiNii Article

researchmap

Language：Japanese   Publisher：The Institute of Image Information and Television Engineers  

DOI： 10.11485/itetr.32.54.0_17

CiNii Article

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

In this paper, we propose a method for composing templates of lung sound classification. First, we obtain a sequence of power spectra by FFT for each given lung sound and compute a small number of component spectra by ICA for each of the overlapping sets of tens of consecutive power spectra. Second, we put component spectra obtained from various lung sounds into a single set and conduct clustering a large number of times. When component spectra belong to the same cluster in all clustering results, these spectra show robust similarity. Therefore, we can use such spectra to compose a template of lung sound classification.

DOI： 10.1007/978-3-540-68125-0_100

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

In this paper, we compare latent Dirichlet allocation (LDA) with probabilistic latent semantic indexing (pLSI) as a dimensionality reduction method and investigate their effectiveness in document clustering by using real-world document sets. For clustering of documents, we use a method based on multinomial mixture, which is known as an efficient framework for text mining. Clustering results are evaluated by F-measure, i.e., harmonic mean of precision and recall. We use Japanese and Korean Web articles for evaluation and regard the category assigned to each Web article as the ground truth for the evaluation of clustering results. Our experiment shows that the dimensionality reduction via LDA and pLSI results in document clusters of almost the same quality as those obtained by using original feature vectors. Therefore, we can reduce the vector dimension without degrading cluster quality. Further, both LDA and pLSI are more effective than random projection, the baseline method in our experiment. However, our experiment provides no meaningful difference between LDA and pLSI. This result suggests that LDA does not replace pLSI at least for dimensionality reduction in document clustering.

DOI： 10.1007/978-3-540-78159-2_2

researchmap

Language：Japanese   Publishing type：Research paper (scientific journal)  

researchmap

Language：Japanese   Publishing type：Research paper (scientific journal)   Publisher：Information Processing Society of Japan (IPSJ)  

The naive Bayes classifier is a well-known method for document classification. However, the naive Bayes classifier gives a satisfying classification accuracy only after an appropriate tuning of the smoothing parameter. Moreover, we should find appropriate parameter values separately for different document sets. In this paper, we focus on an effective probabilistic framework for document classification, called Dirichlet mixtures, which requires no parameter tuning and provides satisfying classification accuracies with respect to various document sets. Many researches in the field of image processing and of natural language processing utilize Dirichlet mixtures. Especially, in the field of natural language processing, many experiments are conducted by using real document data sets. However, most researches use the perplexity as an evaluation measure. While the perplexity is a purely theoretical measure, the accuracy is popular for document classification in the field of information retrieval or of text mining. The accuracy is computed by comparing correct labels with predictions made by the classifier. In this paper, we conduct an evaluation experiment by using 20 newsgroups data set and the Korean Web newspaper articles under the intention that we will use Dirichlet mixtures for multilingual applications. In the experiment, we compare the naive Bayes classifier with the classifier based on Dirichlet mixtures and clarify their qualitative and quantitative differences.

CiNii Article

researchmap

Other Link： http://hdl.handle.net/10069/16317

Language：Japanese   Publishing type：Research paper (scientific journal)  

researchmap

Language：Japanese   Publishing type：Research paper (scientific journal)  

researchmap

Language：Japanese   Publishing type：Research paper (scientific journal)   Publisher：Inst. of Image Information and Television Engineers  

The purpose of our research is to develop a method for recognizing abnormal lung sounds without the need of a medical specialist. Listening to the sounds of the human body is one of the most important methods of checking someone's health. However, identification of abnormal lung sounds is difficult for an untrained person. We differentiated true abnormal sounds from interfering noise by investigating the fact that lung sounds are generated periodically in relation to the breathing cycle.

DOI： 10.3169/itej.61.1769

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：ASSOC COMPUTING MACHINERY  

Results of queries by personal names often contain documents related to several people because of the namesake problem. In order to differentiate documents related to different people, an effective method is needed to measure document similarities and to find documents related to the same person. Some previous researchers have used the vector space model or have tried to extract common named entities for measuring similarities. We propose a new method that uses Web directories as a knowledge base to find shared contexts in document pairs and uses the measurement of shared contexts to determine similarities between document pairs. Experimental results show that our proposed method outperforms the vector space model method and the named entity recognition method.

DOI： 10.1145/1244002.1244188

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/sac/sac2007.html#conf/sac/VuMTA07

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Results of queries by personal names often contain documents related to several people because of the namesake problem. In order to differentiate documents related to different people, an effective method is needed to measure document similarities and to find documents related to the same person. Some previous researchers have used the vector space model or have tried to extract common named entities for measuring similarities. We propose a new method that uses Web directories as a knowledge base to find shared contexts in document pairs and uses the measurement of shared contexts to determine similarities between document pairs. Experimental results show that our proposed method outperforms the vector space model method and the named entity recognition method. © 2007 IEEE.

DOI： 10.1109/RIVF.2007.369155

Scopus

researchmap

Publisher：CID  

researchmap

Publisher：ACM  

DOI： 10.4108/infoscale.2007.203

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

In this paper, we target on the problem of personal name disambiguation in search results returned by personal name queries. Usually, a personal name refers to several people. Therefore, when a search engine returns a set of documents containing that name, they are often relevant to several individuals with the same namesake. Automatic differentiation of people in the resulting documents may help users to search for the person of interest easier. We propose a method that uses web directories to improve the similarity measurement in personal name disambiguation. We carried out experiments on real web documents in which we compared our method with the vector space model method and the named entity recognition method. The results show that our method has advantages over these previous methods. © 2007 IEEE.

DOI： 10.1109/AINAW.2007.367

Scopus

researchmap

Language：Japanese   Publishing type：Research paper (scientific journal)  

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：SPRINGER-VERLAG BERLIN  

This paper explores techniques that discover terms to replace given query terms from a selected subset of documents. The Internet allows access to large numbers of documents archived in digital format. However, no user can be an expert in every field, and they trouble finding the documents that suit their purposes experts when they cannot formulate queries that narrow the search to the context they have in mind. Accordingly, we propose a method for extracting terms from searched documents to replace user-provided query terms. Our results show that our method is successful in discovering terms that can be used to narrow the search.

DOI： 10.1007/11875581_108

researchmap

Language：English  

researchmap

Language：Japanese   Publishing type：Research paper (conference, symposium, etc.)  

CiNii Article

researchmap

Language：Japanese   Publishing type：Research paper (conference, symposium, etc.)  

researchmap

Publisher：National Institute of Informatics (NII)  

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/ntcir/ntcir2005.html#conf/ntcir/MasadaKTA05

Language：English   Publishing type：Research paper (scientific journal)   Publisher：IEICE-INST ELECTRONICS INFORMATION COMMUNICATIONS ENG  

We propose a novel method for Web page grouping based only on hyperlink information. Because of the explosive growth of the World Wide Web, page grouping is expected to provide a general grasp of the Web for effective information search and netsurfing. The Web can be regarded as a gigantic digraph where pages are vertices and links are arcs. Our grouping method is a generalization of decomposition into strongly connected components, in which each group is constructed as a subset of a strongly connected component. Moreover, group sizes can be controlled by adjusting a parameter, called the threshold parameter. We call the resulting groups parameterized connected components (PCCs). The algorithm is simple and admits parallelization. Notably, we apply Dijkstra's shortest path algorithm in our grouping method. This paper also includes experimental results for 15 million Web pages, which show the contribution of our method to efficient Web surfer navigation.

researchmap

Other Link： http://dblp.uni-trier.de/db/journals/ieicet/ieicet87d.html#journals/ieicet/MasadaTA04

Publisher：National Institute of Informatics (NII)  

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/ntcir/ntcir2004.html#conf/ntcir/KanazawaMTA04

Language：English   Publishing type：Research paper (scientific journal)   Publisher：SPRINGER-VERLAG BERLIN  

We propose a novel method for Web page grouping based only on hyperlink information. Because of the explosive growth of the Web, page grouping is expected to provide a general grasp of the Web for effective Web search and netsurfing. The Web can be regarded as a gigantic digraph where pages are vertices and links are arcs. Our method is a generalization of the decomposition into strongly connected components. Each group is constructed as a subset of a strongly connected component. Moreover, group sizes can be controlled by a parameter, called the threshold parameter. We call the resulting groups parameterized connected components. The algorithm is simple and admits parallelization. Notably, we apply Dijkstra's shortest path algorithm in our method.

DOI： 10.1007/978-3-540-24571-1_34

researchmap

Language：Japanese   Publishing type：Research paper (conference, symposium, etc.)  

researchmap

Language：Japanese   Publishing type：Research paper (conference, symposium, etc.)  

researchmap

Publisher：ACM  

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/compgeom/compgeom96.html#conf/compgeom/OnoKMHSNIIIA96

Publisher：ACM  

CiNii Article

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/compgeom/compgeom96.html#conf/compgeom/MasadaII96

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

Demands for more steady and more efficient direct voltage power supplies have been increasing in the context of energy conservation measures. Digital DC-DC converters are especially gathering attention because they have a high degree of reliability and flexibility. This paper addresses an FPGA-based control mechanism for a DC-DC converter. This approach enables programmability for a wide range of control algorithms while higher operational speed is anticipated. In this paper, we discuss required arithmetic precision and trade-off relationship between control accuracy and hardware costs through prototype implementation of an FPGA-based DC-DC converter. Evaluation of the prototype system reveals that fixed point arithmetic with 10-bit fraction part is enough in terms of dynamic characteristics. Design of a counter module which uses multiple phase-shifted clock signals to increase the PWM resolution while keeping the system clock frequency low is also discussed.

CiNii Article

researchmap

Language：English   Publisher：The Institute of Electronics, Information and Communication Engineers  

Reconfigurable systems are known to be able to achieve higher performance than traditional microprocessor architecture for many application fields. However, in order to extract a full potential of the reconfigurable systems, programmers often have to design and describe the best suited code for their target architecture with specialized knowledge. The aim of this paper is to assist the users of reconfigurable systems by implementing a programming environment with a multithread model. The experimental results show our translator automatically generates efficient performance-aware code segments including DMA transfer and shift registers for memory access optimization.

CiNii Article

researchmap

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

CiNii Article

researchmap

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

In this paper, we propose a GPU acceleration of multi-topic extraction from images by using LDA (latent Dirichlet allocation). LDA is originally proposed as a probabilistic model for documents by Blei et al. In recent days, LDA is applied to multimedia information other than documents. We provide the results of experiments where we apply LDA to Professor Wang's 10,000 test images and extract multiple visional topics. We adpot collapsed variational Bayesian inference method for LDA and accelerate this by using Nvidia CUDA compatible GPU devices.

CiNii Article

researchmap

Language：Japanese   Publisher：Information Processing Society of Japan (IPSJ)  

In this paper, we employ the latent Dirichlet allocation as a method for the dimensionality reduction of feature vectors and reveal its effectiveness in document clustering. In the evaluation experiment, we perform clustering on the document sets of Japanese and Korean Web news articles. We regard the categories assigned to each article as the ground truth of clustering evaluation. We compare the clustering results obtained by using the feature vectors whose entries are term frequencies with the results obtained by using the feature vectors whose dimensions are reduced by the latent Dirichlet allocation.

CiNii Article

researchmap

Other Link： http://id.nii.ac.jp/1001/00018810/

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

In this paper, we employ the latent Dirichlet allocation as a method for the dimensionality reduction of feature vectors and reveal its effectiveness in document clustering. In the evaluation experiment, we perform clustering on the document sets of Japanese and Korean Web news articles. We regard the categories assigned to each article as the ground truth of clustering evaluation. We compare the clustering results obtained by using the feature vectors whose entries are term frequencies with the results obtained by using the feature vectors whose dimensions are reduced by the latent Dirichlet allocation.

CiNii Article

researchmap

researchmap

Language：English   Publisher：Information Processing Society of Japan (IPSJ)  

Results of queries by personal names often contain documents related to several people because of namesake problem. In order to discriminate documents related to different people, it is required an effective method to measure document similarities and to find out relevant documents of the same person. Some previous researches have used cosine similarity method or have tried to extract common named entities for measuring similarities. We propose a new method which uses web directories as knowledge base to find out shared contexts in document pairs and uses the measurement of shared contexts as similarities between document pairs. Experimental results show that our proposed method outperforms cosine similarity method and common named entities method.

CiNii Article

researchmap

Other Link： http://id.nii.ac.jp/1001/00018907/

Language：English   Publisher：The Institute of Electronics, Information and Communication Engineers  

Results of queries by personal names often contain documents related to several people because of namesake problem. In order to discriminate documents related to different people, it is required an effective method to measure document similarities and to find out relevant documents of the same person. Some previous researches have used cosine similarity method or have tried to extract common named entities for measuring similarities. We propose a new method which uses web directories as knowledge base to find out shared contexts in document pairs and uses the measurement of shared contexts as similarities between document pairs. Experimental results show that our proposed method outperforms cosine similarity method and common named entities method.

CiNii Article

researchmap

researchmap

researchmap

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

Conventional search engines are designed mainly for general keyword search. Therefore, in many cases, we can find no appropriate combination of query terms. In this paper, we present a query disambiguation method by using infrequent term cooccurrences. This strategy comes from the following idea : terms appearing with a wide variety of terms cannot establish an independent topic. Based on this hypothesis, terms are weighted. The experimental results show that the terms ranked higher by our method can improve the average precision of Web search when added to the original query terms. As compared with other term ranking methods, our method gives higher ranks to the terms denoting more particular and adequate stuff and referring to more specific concepts.

CiNii Article

researchmap

researchmap

researchmap

Language：English   Publisher：The Institute of Electronics, Information and Communication Engineers  

We propose a novel method for Web page grouping based only on hyperlink information. Because of the explosive growth of the World Wide Web, page grouping is expected to provide a general grasp of the Web for effective information search and netsurfing. The Web can be regarded as a gigantic digraph where pages are vertices and links are arcs. Our grouping method is a generalization of decomposition into strongly connected components, in which each group is constructed as a subset of a strongly connected component. Moreover, group sizes can be controlled by adjusting a parameter, called the threshold parameter. We call the resulting groups parameterized connected components (PCCs). The algorithm is simple and admits parallelization. Notably, we apply Dijkstra's shortest path algorithm in our grouping method. This paper also includes experimental results for 15 million Web pages, which show the contribution of our method to efficient Web surfer navigation.

CiNii Article

researchmap

Language：Japanese   Publisher：日本データベース学会  

CiNii Article

researchmap

researchmap

Language：English   Publisher：WORLD SCIENTIFIC PUBL CO PTE LTD  

We propose algorithms to enumerate (1) regular triangulations, (2) spanning regular triangulations, (3) equivalence classes of regular triangulations with respect to symmetry, and (4) all triangulations. All of the algorithms are for arbitrary points in general dimension. They work in output-size sensitive time with memory only of several times the size of a triangulation. For the enumeration of regular triangulations, we use the fact by Gel'fand, Zelevinskii and Kapranov that regular triangulations correspond to the vertices of the secondary polytope. We use reverse search technique by Avis and Fukuda, its extension for enumerating equivalence classes of objects, and a reformulation of a maximal independent set enumeration algorithm. The last approach can be extended for enumeration of dissections.

DOI： 10.1142/S0218195902000980

researchmap

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

The rapid growth of the amount of information on WWW makes Web search methods based only on textual information more and more unrealistic. In recent years, many researches provide attractive link-based retrieving methods. This paper proposes a method for link-based Web page grouping, which aims to reduce the complexity of following text-based retrievals by enlarging the size of units for those retrievals. This method also makes the granularity of groups controllable by adjusting one threshold parameter. This paper includes the results of preliminary experiments, which clarify the characteristic of proposed grouping method.

CiNii Article

researchmap

Language：Japanese   Publisher：日本データベース学会  

CiNii Article

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

Presentation type：Poster presentation  

researchmap

researchmap

researchmap

researchmap

Presentation type：Poster presentation  

researchmap

Presentation type：Poster presentation  

researchmap

researchmap

researchmap

Presentation type：Poster presentation  

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap