著者
中村 友昭 長井 隆行 船越 孝太郎 谷口 忠大 岩橋 直人 金子 正秀
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.3, pp.498-509, 2015-05-01 (Released:2015-03-26)
参考文献数
30
被引用文献数
1 3

Humans develop their concept of an object by classifying it into a category, and acquire language by interacting with others at the same time. Thus, the meaning of a word can be learnt by connecting the recognized word and concept. We consider such an ability to be important in allowing robots to flexibly develop their knowledge of language and concepts. Accordingly, we propose a method that enables robots to acquire such knowledge. The object concept is formed by classifying multimodal information acquired from objects, and the language model is acquired from human speech describing object features. We propose a stochastic model of language and concepts, and knowledge is learnt by estimating the model parameters. The important point is that language and concepts are interdependent. There is a high probability that the same words will be uttered to objects in the same category. Similarly, objects to which the same words are uttered are highly likely to have the same features. Using this relation, the accuracy of both speech recognition and object classification can be improved by the proposed method. However, it is difficult to directly estimate the parameters of the proposed model, because there are many parameters that are required. Therefore, we approximate the proposed model, and estimate its parameters using a nested Pitman--Yor language model and multimodal latent Dirichlet allocation to acquire the language and concept, respectively. The experimental results show that the accuracy of speech recognition and object classification is improved by the proposed method.
著者
中林 一貴 益井 博史 谷口 忠大
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.35, no.5, pp.G-K31_1-10, 2020-09-01 (Released:2020-09-01)
参考文献数
19

Communication-field mechanism design is to design a mechanism including rules and incentives to indirectly control a group of people having communication, e.g., discussion, debate, meeting, and consultation. A communication-field mechanism is expected to give constraints to the actual communications. We hypothesized that such constraints are beneficial for the application of technologies based on artificial intelligence. In this paper, we evaluate this concept by taking an automatic speech recognition system and dealing rights to speak (DRS) as an example of proof of concept. An experiment shows that the simple introduction of DRS improves the performance of speech recognition.
著者
益井 博史 海川 由美子 三苫 奈美子 谷口 忠大
出版者
一般社団法人 システム制御情報学会
雑誌
システム制御情報学会論文誌 (ISSN:13425668)
巻号頁・発行日
vol.32, no.12, pp.439-445, 2019-12-15 (Released:2020-03-15)
参考文献数
17

This paper analyzes the degree of influence of the presentation order on the voting behavior of participants in the book review game, Bibliobattle, and examines ways to reduce this influence. We collected and analyzed the data of Bibliobattle games that were spontaneously conducted in various places, and this method was applied to research other communication-field mechanism designs. We classified the results of approximately 800 Bibliobattle games collected from the Internet by the order of presentation. Subsequently, we compared the number of Champion Book awards secured by the first and second, and last and last but one in the presentations when compared to others in the presentation order. Consequently, the possibility of the first and second presenters acquiring a Champion Book award has a detrimental effect on the other presenters in the order. Conversely, the possibility of the last and second-last presenters acquiring a Champion Book award is advantageous for the other presenters in the order. We considered the possibility that the response order effect influences voting in the Bibliobattle game. Also, by performing the voting process in the reverse order of the presentation, we examined the way to reduce the influence of the response order effect.
著者
林 楓 岩田 具治 谷口 忠大
出版者
一般社団法人 人工知能学会
雑誌
人工知能学会全国大会論文集 第32回全国大会(2018)
巻号頁・発行日
pp.4A104, 2018 (Released:2018-07-30)

クラスタリングは機械学習および人工知能の分野において重要なタスクである.確率論的生成モデルは効率的な推論のためにはデータに対して強い仮定が必要であり,混合ガウスモデル(GMM)を用いたクラスタリングには特徴エンジニアリングが必要であった.ここ数年,Variational Autoencoder(VAE)とGMMを組み合わせたモデルで複雑なデータをクラスタリングする研究が注目されている.本稿では深層混合モデル(DMM)を提案する. DMMでは,まず潜在的なベクトルがGMMにより生成され、次に潜在ベクトルが観測データに変換される.DMMは結合尤度の下限を最大化することで訓練される.実験では,提案モデルは,GMMによってクラスタリングすることが困難なデータのベースラインの手法と比較して最も良い性能を示した.