著者

Yuanzhi Ke Masafumi Hagiwara

出版者

The Japanese Society for Artificial Intelligence

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.33, no.4, pp.D-I23-1-8, 2018-07-01 (Released:2018-07-02)

参考文献数

25

被引用文献数

3

---

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.

著者

大谷 紀子 志村 正道

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.19, no.5, pp.399-404, 2004 (Released:2004-07-06)

参考文献数

11

被引用文献数

2 1

---

In representing classification rules by decision trees, simplicity of tree structure is as important as predictive accuracy especially in consideration of the comprehensibility to a human, the memory capacity and the time required to classify. Trees tend to be complex when they get high accuracy. This paper proposes a novel method for generating accurate and simple decision trees based on symbiotic evolution. It is distinctive of symbiotic evolution that two different populations are evolved in parallel through genetic algorithms. In our method one's individuals are partial trees of height 1, and the other's individuals are whole trees represented by the combinations of the former individuals. Generally, overfitting to training examples prevents getting high predictive accuracy. In order to circumvent this difficulty, individuals are evaluated with not only the accuracy in training examples but also the correct answer biased rate indicating the dispersion of the correct answers in the terminal nodes. Based on our method we developed a system called SESAT for generating decision trees. Our experimental results show that SESAT compares favorably with other systems on several datasets in the UCI repository. SESAT has the ability to generate more simple trees than C5.0 without sacrificing predictive accuracy.

著者

中小路 久美代 山本 恭裕

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.19, no.2, pp.154-165, 2004 (Released:2004-01-27)

参考文献数

41

被引用文献数

3 3

---

This paper describes our approach for the development of application systems for creative knowledge work, particularly for early stages of information design tasks. Being a cognitive tool serving as a means of externalization, an application system affects how the user is engaged in the creative process through its visual interaction design. Knowledge interaction design described in this paper is a framework where a set of application systems for different information design domains are developed based on an interaction model, which is designed for a particular model of a thinking process. We have developed two sets of application systems using the knowledge interaction design framework: one includes systems for linear information design, such as writing, movie-editing, and video-analysis; the other includes systems for network information design, such as file-system navigation and hypertext authoring. Our experience shows that the resulting systems encourage users to follow a certain cognitive path through graceful user experience.

著者

立花 誠人 村田 剛志

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.34, no.5, pp.B-IC2_1-8, 2019-09-01 (Released:2019-09-01)

参考文献数

14

被引用文献数

1

---

Since several types of data can be represented as graphs, there has been a demand for generalizing neural network models for graph data. Graph convolution is a recent scalable method for performing deep feature learning on attributed graphs by aggregating local node information over multiple layers. Such layers only consider attribute information of node neighbors in the forward model and do not incorporate knowledge of global network structure in the learning task. In this paper, we present a scalable semi-supervised learning method for graph-structured data which considers not only neighbors information, but also the global network structure. In our method, we add a term preserving the network structural features such as centrality to the objective function of Graph Convolutional Network and train for both node classification and network structure preservation simultaneously. Experimental results showed that our method outperforms state-of-the-art baselines for the node classification tasks in the sparse label regime.

著者

辻 順平 能登 正人

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.33, no.2, pp.C-H98_1-9, 2018-03-01 (Released:2018-03-01)

参考文献数

14

---

In the theme park problem, it is important to find a coordination algorithm that effectively shortens the visiting time of an entire theme park while guaranteeing individual optimality for each visitor. In a previous study, a coordination algorithm, called statement-based cost estimate (SCE), was proposed that allows individual visitors to select plans that minimize a visitor’s visiting time while shortening the visiting time of the entire theme park. However, the improvement in visiting time was not sufficient from their experiment using SCE. We thought it necessary to relax the premise constraint “minimize individual visiting time” to further improve SCE. In this paper, we propose a framework to further reduce visiting time by considering Pareto optimality. In the proposed framework, each visitor determines several Pareto optimal plans based on the evaluation value calculated using SCE and presents them to a coordination system. Then, the coordination system searches for the entire optimal plan that minimizes the predicted value of the total visiting time of the entire theme park among the Pareto optimal plan candidates. The proposed framework guarantees visitors’ “personal optimality” in the meaning of Pareto optimality, and there is a possibility that the framework will shorten the visiting time of the entire theme park. We conducted a simulation experiment using a coordination algorithm based on the proposed framework and clarified the effectiveness of the framework.

著者

浜田 良樹 松尾 徳朗 伊藤 孝行

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.33, no.5, pp.A-I13_1-4, 2018-09-01 (Released:2018-09-03)

参考文献数

10

---

The authors’ group has been developed AI-based Online Consensus Building Support System called COLLAGREE and used around our university since 2010. The goal of our research is to spread this system to other local governments. However, new technology which directly concerns to society might make a conflict with current society. This problem has been commonly known as ELSI. In this paper, we analyzed our past experiments by using our system, developed the reason of success, and ELSI problems that need to solve. To achieve this purpose, we launched ELSI Committee on December 2017. Our approach should be useful for other AI-based systems, when researcher has been conscious on ELSI issue.

著者

藁谷 敏晴

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.20, no.6, pp.448-460, 2005 (Released:2005-10-12)

参考文献数

27

被引用文献数

2 2

---

The main aim of this paper is to propose an appropriate logical system that is suitable to describe the notion of IS-A link as well as is-a link. The most important point to be realized is that those relations are not set thoretical ones. They connect two `general names' to construct a proposition, so that what is needed for proper descriptions of the relations in question is a theory of general names. It will be shown that is-a is a logical unit of axiomatically determined behaviour. The axiom concerning is-a relation was established by S. Lesniewski who named his theory of general names ontology. Today `ontology' has also become a common term for AI researchers. I intend to make it clear that there is a close connection between `ontology' used by Lesniewski and by AI researchers, even though they developed quite independently. I wish to stress that ontology created by Lesniewski is a system of syllogistic equipped with singular propositions and the theory of quantification. To make this point clear, I proposed a fragment of syllogism that I called MO(minimal ontology). This paper includes comments and examples articulating the logical power of ontology.

著者

藤原 靖宏 中辻 真 塩川 浩昭 三島 健 鬼塚 真

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.30, no.2, pp.473-478, 2015-03-01 (Released:2015-02-03)

参考文献数

6

---

In AI communities, many applications utilize PageRank. To obtain high PageRank score nodes, the original approach iteratively computes the PageRank score of each node until convergence from the whole graph. If the graph is large, this approach is infeasible due to its high computational cost. The goal of this study is to find top-k PageRank score nodes efficiently for a given graph without sacrificing accuracy. Our solution, F-Rank, is based on two ideas: (1) It iteratively estimates lower/upper bounds of PageRank scores, and (2) It constructs subgraphs in each iteration by pruning unnecessary nodes and edges to identify top-k nodes. Experiments show that F-Rank finds top-k nodes much faster than the original approach.

著者

井手 一郎 木下 智義 高橋 友和 孟 洋 片山 紀生 佐藤 真一 村瀬 洋

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.23, no.5, pp.282-292, 2008

被引用文献数

9 2

---

Recent increase of digital storage capacity has enabled the creation of large-scale on-line broadcast video archives. In order to make full use of the data in the archive, it is necessary to let a user easily grasp the availability of certain video data and their contents. Considering this problem, we have been investigating efficient and effective retrieval and reusing methodologies of archived video data. The archive used as a test-bed consists of more than 1,000 hours of news video obtained from a Japanese news program during the past six years. This paper first proposes a news topic tracking and structuring method. A structure called the `topic thread structure', is organized so that it should represent the temporal flow of news topics originating from a specified news story. The paper next introduces a browsing and editing interface that enables the user to browse through news stories along the topic thread structure, and also assists the compilation of selected news stories as a customized video summary or a documentary. The method was applied to the archived news video data in order to observe the quality of the topic thread structure and the usability of the prototype interface. As a result, some structures represented the flow of topics quite close to real-world comprehension. In addition, experiments showed that when the structure could be considered meaningful, the interface combined with the structure could drastically reduce the time needed to browse through the archive for news stories related to the user's interest.

著者

寺田 和憲 山田 誠二 伊藤 昭

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.27, no.2, pp.73-81, 2012

---

We conducted an experimental investigation on human adaptation to change in an agent's strategy through a competitive two-player game. Modeling the process of human adaptation to agents is important for designing intelligent interface agents and adaptive user interfaces that learn a user's preferences and behavior strategy. However, few studies on human adaptation to such an agent have been done. We propose a human adaptation model for a two-player game. We prepared an on-line experimental system in which a participant and an agent play a repeated penny-matching game with a bonus round. We then conducted experiments in which different opponent agents (human or robot) change their strategy during the game. The experimental results indicated that, as expected, there is an adaptation phase when a human is confronted with a change in the opponent agent's strategy, and adaptation is faster when a human is competing with robot than with another human.

著者

稲葉 通将 高橋 健一

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.34, no.2, pp.E-I94_1-9, 2019-03-01 (Released:2019-03-01)

参考文献数

28

---

Non-task-oriented dialogue systems are required to chat with users in accordance with their interests. In this study, we propose a neural network-based method for estimating speakers’ levels of interest from dialogues. Our model first converts given utterances into utterance vectors using a word sequence encoder with word attention. Afterward, our novel attention approach, sentence-specific sentence attention extracts useful information for estimating the level of interest. Additionally, we introduce a new pre-training method for our model. Experimental results indicated that it was most effective to use topic-specific sentence attention and proposed pre-training in combination.

著者

高津 弘明 福岡 維新 藤江 真也 岩田 和彦 小林 哲則

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.34, no.2, pp.B-I65_1-15, 2019-03-01 (Released:2019-03-01)

参考文献数

46

被引用文献数

1

---

We have been developing a speech-based “news-delivery system”, which can transmit news contents via spoken dialogues. In such a system, a speech synthesis sub system that can flexibly adjust the prosodic features in utterances is highly vital: the system should be able to highlight spoken phrases containing noteworthy information in an article; it should also provide properly controlled pauses between utterances to facilitate user’s interactive reactions including questions. To achieve these goals, we have decided to incorporate the position of the utterance in the paragraph and the role of the utterance in the discourse structure into the bundle of features for speech synthesis. These features were found to be crucially important in fulfilling the above-mentioned requirements for the spoken utterances by the thorough investigation into the news-telling speech data uttered by a voice actress. Specifically, these features dictate the importance of information carried by spoken phrases, and hence should be effectively utilized in synthesizing prosodically adequate utterances. Based on these investigations, we devised a deep neural network-based speech synthesis model that takes as input the role and position features. In addition, we designed a neural network model that can estimate an adequate pause length between utterances. Experimental results showed that by adding these features to the input, it becomes more proper speech for information delivery. Furthermore, we confirmed that by inserting pauses properly, it becomes easier for users to ask questions during system utterances.

著者

來村 徳信 笠井 俊信 吉川 真理子 高橋 賢 古崎 晃司 溝口 理一郎

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.17, no.1, pp.73-84, 2002 (Released:2002-04-04)

参考文献数

19

被引用文献数

4 4

---

In conceptual design, a designer decomposes a required function into sub-functions, so-called functional decomposition, using a kind of functional knowledge representing achievement relations among functions. Aimin at systematization of such functional knowledge, we proposed ontologies that guide conceptualization of artifacts from the functional point of view. This paper discusses its systematic description based on the functional ontologies. Firstly, we propose a new concept named “way of achievement” as a key concept for its systematization. Categorization of typical representations of the knowledge and organization as is-a hierarchies are also discussed. Such concept, categorization, and functional ontologies make the functional knowledge consistent and applicable to other domains. Next, the implementation of the functional ontologies and their utility on description of the knowledge are shown. Lastly, we discuss development of a knowledge-based system to help human designers redesign an existin artifact. The ontology of functional concepts and the systematic description of functional knowledge enable the supporting system to show designers a wide range of alternative ways and then to facilitate innovative redesign.

著者

石橋 賢 深瀧 創 宮田 一乘

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.30, no.1, pp.229-236, 2015

被引用文献数

4

---

Recently, sizzle words have been utilized for various product packages. These words have effective communicative performance to convey deliciousness of food. In particular, onomatopoeias are used for many product packages because they can convey the texture of food sensuously. When using sizzle words, producers may consider consumers' impressions. This study aims to investigate the relations between consumers' impressions and sizzle words through an experiment. The experimental targets were four rice crackers of different hardness because their textures were directly related to consumers' impressions and some packages of rice crackers used a few sizzle words for advertising. The results of this experiment indicated five tendencies: (1) the sizzle words regarding satisfaction, tradition, and typicality, and the onomatopoeias including "<i>zaku</i>" are appropriate for an extremely hard rice cracker; (2) the sizzle words regarding lightness and comfort, and the onomatopoeias including "<i>saku</i>" are appropriate for a non-hard rice cracker; (3) the sizzle words regarding aridity, fineness, and unforgettable taste are appropriate for a slightly hard rice cracker; (4) the onomatopoeias expressing crunchy texture are appropriate for a quite hard rice cracker; (5) rough texture of food brings a sense of satiety. In addition, we performed a factor analysis using the results with 14 onomatopoeias in the experiment. The analysis results showed three factors: brittleness, irritation, and lightness. In future, we expect that these results can be utilized for guiding a choice of an appropriate sizzle word.

著者

松中 春樹 丹生 智也 番原 睦則 田村 直之

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.27, no.2, pp.10-15, 2012 (Released:2012-01-13)

参考文献数

15

---

Propositional Satisfiability (SAT) is fundamental in solving many application problems in Artificial Intelligence and Computer Science. Remarkable improvements in the efficiency of SAT solvers have beenmade over the last decade. Such improvements encourage researchers to solve constraint satisfaction problems by encoding them into SAT (i.e. ``SAT encoding''). Balanced Incomplete Block Design (BIBD) is one of the most typical block designs. BIBDs have been applied in several fields such as design experiments, coding theory, and cryptography. In this paper, we consider the problem of generating BIBDs by SAT encoding. We present a new SAT encoding that is an enhancement of order encoding with the idea of binary tree. It is designed to reduce the number of clauses required for cardinality constraints, compared with order encoding. In our experiments, our encoding was able to give a greater number of solutions than order encoding and state-of-the-art constraint solvers Mistral and choco.

著者

出口 幸子 白井 克彦

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.18, no.3, pp.153-160, 2003 (Released:2003-04-08)

参考文献数

16

---

This paper describes the knowledge representation of the melody and rhythm in koto songs based on the structure of the domain: the scale, melisma (the melody in a syllable), and bar. We have encoded koto scores and extracted 2,3,4-note melodic patterns sequentially from the voice part of koto scores. The 2,3,4-note patterns used in the melisma are limited and the percentages of top patterns are high. The 3,4-note melodic patterns are examined at each scale degree. These patterns are more restricted than the patterns that are possible under the constraint of the scale. These typical patterns on the scale represent the knowledge of koto players. We have analyzed rhythms in two different ways. We have extracted rhythms depending on each melodic pattern, while we have extracted rhythms depending on each bar. The former are complicated and the latter are typical. This result indicates that koto players recognize melodic patterns and rhythmic patterns independently. Our analyses show the melodic patterns and rhythmic patterns that are acquired by koto players. These patterns will be applied to the description of variations of the melisma to build a score database. These patterns will also be applied to a composition and education. The melodic patterns can be extracted from other genres of Japanese traditional music, foreign old folk songs or chants by using this method.

著者

熊本 忠彦 太田 公子

出版者

The Japanese Society for Artificial Intelligence

雑誌

人工知能学会論文誌 = Transactions of the Japanese Society for Artificial Intelligence : AI (ISSN:13460714)

巻号頁・発行日

vol.21, pp.310-318, 2006-11-01

参考文献数

12

被引用文献数

7 5

---

Impression-based music retrieval helps users in finding musical pieces that suit their preferences, feelings, or mental states from the huge volume of a music database. We have therefore developed an impression-based music retrieval system that enables this. Users are asked to select one or more pairs of impression words from the multiple pairs presented by the system and estimate each of the selected pairs on a seven-step scale in order to input their impressions into the system. For instance, if they want to locate musical pieces that will create a happy impression, they should check the radio button ``Happy'' in the impression scale, ``Very happy -- Happy -- A little happy -- Neutral -- A little sad -- Sad -- Very sad,'' where a pair of impression words with a seven-step scale is called an ``impression scale'' in this paper. The system would measure the distance between the impressions of every musical piece in a user-specified music database and the impressions inputted by the user, and determine candidate musical pieces to be presented as retrieval results. In this paper, we define the form of vectors that numerically express impressions of musical pieces, and propose a method of generating such a vector from a musical piece. The most significant attribute of this method is that it uses n-gram statistics of information on pitch, strength, and length of every tone in that musical piece as features extracted from it. We also present the results of evaluating the performance of the system.

著者

柳川 由紀子 越後 富夫 宮崎 祐太 武村 紀子 八木 康史

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.33, no.6, pp.C-I33_1-12, 2018-11-01 (Released:2018-11-01)

参考文献数

51

---

Tracking precisely of abnormalities in the gastrointestinal tract is useful for preparing sample image sequences on educational training for medical diagnose on endoscopy. While the gastrointestinal wall deforms continuously in an unpredictable manner, however, abnormalities without distinctive features make it difficult to track over continuous frames. To address this problem, the proposed method employs Convolutional neural networks (CNN) for tracking lesion area. Conventionally, CNN for tracking requires a large amount of sample data for preliminary learning. The state-of-arts tracking methods using CNN are premised on preliminary learning on data similar to target images given a large number of correct answer labels. On the other hand, the proposed method are not required preliminary learning using similar data. The image components in the marked region at the starting frame is similar to components at the only same position, but different between them depending on the degree of overlapped area. Furthermore, in the successive frame, the components in the previous region is similar to them in the identified area. Therefore, similarity can be learned in the previous frame, called it as an intra-frame training. This paper describes the method for tracking an abnormal region by using CNN based on training overlap rates between the abnormal region and local scanning one with the same size on the starting intra-frame. Furthermore, network parameters are transformed from training the similar regions on the continuous frame additionally. We demonstrate the efficiency of the proposed approach using eight common types of gastrointestinal abnormality.

著者

郷古 学 金 天海

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.32, no.5, pp.E-H31_1-8, 2017-09-01 (Released:2017-09-01)

参考文献数

26

被引用文献数

1

---

Keeping a living and working spaces tidy is very important for healthy daily life. The efficiency in working rises when the office is tidy. We developed a system that tidies up through the cooperation between a robot and a human. In this study, we investigated the influence of a robot’s behavior on the motivation of tidying up. For completing this system, it is necessary to investigate the effective behaviors that encourages a human. To validate what behavior effectively encourage human to tidy up, we conducted a preliminary experiment with 13 male and 5 female participants, aged 20-23. We found the statistically significant difference between the cases where the robot took actions or not.

著者

鳥海 不二夫 西岡 寛兼 梅岡 利光 石井 健一郎

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.27, no.3, pp.143-150, 2012 (Released:2012-03-27)

参考文献数

14

被引用文献数

1 2

---

The financial markets are fluctuating consistently. Therefore, it is difficult to analyze the financial market based on the same theory, without depending on the state of the market. So we use the concept ofmarket condition change. To estimate the points when the market change occurred in a real market is effective for market analysis. Thus, in this paper, we propose a method to detect the changes in market conditions. In the proposed method, we focuse on the stock board instead of the price data. From the stock board data, we classify short time series data to clusters by using k-means clustering method. Then, we generate Hidden Markov Model(HMM) from the transition probability of each clusters. By using the likelihood of HMM, we analyze the similarities of each time series data. We performed an experiment to evaluate the effectiveness of the method by discriminant analysis of time series data which created from opening session and continuous session. As a result, two time series data are discriminated with high accuracy. Finally, we compared the discriminate performance of proposed method with another discriminant analysis methods. We used three types of time series data of stock board and price data, before the Lehman's fall financial crisis. From the result, the proposed method shows the best performance in discriminating each financial data.