人工知能学会論文誌
Online ISSN : 1346-8030
Print ISSN : 1346-0714
ISSN-L : 1346-0714
33 巻 , 4 号
選択された号の論文の5件中1~5を表示しています
原著論文
  • 山田 奉子, 上田 洋, 村上 晴美, 岡 育生
    2018 年 33 巻 4 号 p. A-H91_1-13
    発行日: 2018/07/01
    公開日: 2018/07/02
    ジャーナル フリー

    Recently the importance of mathematical information retrieval (MIR) has been recognized and various methods for mathematical expression retrieval have also been proposed. However, since mathematical expressions on the Web are not annotated with natural language, searching for mathematical expressions by conventional search engines is difficult. For helping people in various fields who use mathematics as a learning tool, our proposed method performs a Web search using a mathematical term as a query, extracts mathematical expression images (math-images) related to the query from the obtained Web pages, and presents the top ten math-images with their surrounding information. The method measures the relevance between a query and a math-image from the following viewpoints: the math-image is in a separate line, it has the query in the neighborhood and appropriate image feature quantities, and it appears in the first part of the Web page. We use a support vector machine to discriminate if the image provides appropriate feature quantities. We conducted two experiments with our proposed method. We determined its scoring parameters in Experiment 1 and evaluated it in Experiment 2. The results revealed the usefulness of our proposed method with accuracy, recall, F-measure, mean reciprocal rank (MRR), and mean average precision (MAP).

  • 加藤 大貴, 平山 高嗣, 道満 恵介, 井手 一郎, 川西 康友, 出口 大輔, 村瀬 洋
    2018 年 33 巻 4 号 p. B-HC2_1-9
    発行日: 2018/07/01
    公開日: 2018/07/02
    ジャーナル フリー

    There are few studies on the motion attribute of gaits because of the absence of appropriate semantic labels describing motion. We focus on onomatopoeias to describe the motion of gaits. The Japanese language is known to have a greater number of onomatopoeias in its vocabulary. Especially, the human gait is one of the most commonly represented phenomenon by onomatopoeias expressing its visually dynamic state. It is said that Japanese onomatopoeias have sound-symbolism and their phonemes are strongly related to the impression of various phenomena. Because of this, Japanese people can distinguish gaits based on their appearances and express their impressions on them briefly and intuitively using various onomatopoeias. In addition, a previous study revealed relative body-parts movement is associated with onomatopoeias when we describe gaits. Inspired by these studies, we considered that if a phonetic space based on sound-symbolism can be associated with the kinetic feature space of gaits, subtle difference of gaits could be expressed as difference in phoneme. In this paper, we propose a framework to convert the relative body-parts movements to any onomatopoeia using a regression model. This framework is expected to make human-computer interaction more intuitive. Through experiments, we confirmed the effectiveness of the proposed framework, and discussed the potential of describing an arbitrary gait by not only existing onomatopoeias but also a novel one.

  • Kazuki Yoshiyama, Akito Sakurai
    2018 年 33 巻 4 号 p. C-HA2_1-10
    発行日: 2018/07/01
    公開日: 2018/07/02
    ジャーナル フリー

    The goal of semi-supervised learning is to utilize many unlabeled samples under a situation where a few labeled samples exist. Recently, researches of semi-supervised learning are evolving with deep learning technology development, because, in deep, models have powerful representation to make use of abundant unlabeled samples. In this paper, we propose a novel semi-supervised learning method with uncertainty. It naturally extends the consistency loss under the uncertainty and propose suitable regularizations for the uncertainty. Using two datasets CIFAR-10 and SVHN and with various experiments, we empirically demonstrate that the proposed method achieves competitive or higher performance in accuracy when compared to semi-supervised learning with the conventional consistency loss while our proposal can let a model generalize much faster.

  • Yuanzhi Ke, Masafumi Hagiwara
    2018 年 33 巻 4 号 p. D-I23-1-8
    発行日: 2018/07/01
    公開日: 2018/07/02
    ジャーナル フリー

    Although word embeddings are powerful, weakness on rare words, unknown words and issues of large vocabulary motivated people to explore alternative representations. While the character embeddings have been successful for alphabetical languages, Japanese is difficult to be processed at the character level as well because of the large vocabulary of kanji, written in the Chinese characters. In order to achieve fewer parameters and better generalization on infrequent words and characters, we proposed a model that encodes Japanese texts from the radical-level representation, inspired by the experimental findings in the field of psycholinguistics. The proposed model is comprised of a convolutional local encoder and a recurrent global encoder. For the convolutional encoder, we propose a novel combination of two kinds of convolutional filters of different strides in one layer to extract information from the different levels. We compare the proposed radical-level model with the state-of-the-art word and character embedding-based models in the sentiment classification task. The proposed model outperformed the state-of-the-art models for the randomly sampled texts and the texts that contain unknown characters, with 91% and 12% fewer parameters than the word embedding-based and character embedding-based models, respectively. Especially for the test sets of unknown characters, the results by the proposed model were 4.01% and 2.38% above the word embedding-based and character embedding-based baselines, respectively. The proposed model is powerful with cheaper computational and storage cost, can be used for devices with limited storage and to process texts of rare characters.

  • 海野 一則, 菊地 剛正, 國上 真章 , 山田 隆志, 寺野 隆雄
    2018 年 33 巻 4 号 p. E-HB3_1-9
    発行日: 2018/07/01
    公開日: 2018/07/02
    ジャーナル フリー

    This research has two objectives: (1) to model the momentum effect, (2) to propose a portfolio selection algorithm MESPSA that can use the momentum effect to obtain excess profit. The momentum effect is a phenomenon in which stocks that rise (decline) tend to continue to rise (decline), and momentum effect is a phenomenon often seen in the stock market. However, because existing research does not separate momentum effects from stock price fluctuations it is not always possible to obtain excess return when working with an unknown data set that contains a momentum effect. In this research, we define a new External Force Momentum Effect (EFME) model based on bias in stock price rises (declines). We prepare an artificial data set that contained this momentum effect and construct a portfolio with the proposed algorithm. The relationship between the EFME model and excess return is then analyzed to verify that excess profit can be obtained. Also, we confirm that the proposed algorithm for the actual stock price data set yields excess profits.

feedback
Top