著者

竹内 誉羽 庄野 修 辻野 広司

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.92-102, 2012 (Released:2012-02-10)

参考文献数

23

---

Future robots/agents will perform situated behaviors for each user. Flexible behavioral learning is required for coping with diverse and unexpected users' situations. Unexpected situations are usually not tractable for machine learning systems that are designed for pre-defined problems. In order to realize such a flexible learning system, we were trying to create a learning model that can function in several different kinds of state transitions without specific adjustments for each transition as a first step. We constructed a modular neural network model based on reinforcement learning. We expected that combining a modular architecture with neural networks could accelerate the learning speed of neural networks. The inputs of our neural network model always include not only observed states but also memory information for any transition. In pure Markov decision processes, memory information is not necessary, rather it can lead to lower performance. On the other hand, partially observable conditions require memory information to select proper actions. We demonstrated that the new learning model could actually learn those multiple kinds of state transitions with the same architectures and parameters, and without pre-designed models of environments. This paper describes the performances of constructed models using probabilistically fluctuated Markov decision processes including partially observable conditions. In the test transitions, the observed state probabilistically fluctuated. The new learning model could function in those complex transitions. In addition, the learning speeds of our model are comparable to a reinforcement learning algorithm implemented with a pre-defined and optimized table-representation of states.

著者

山本 俊一 中臺 一博 辻野 広司 奥乃 博

出版者

The Robotics Society of Japan

雑誌

日本ロボット学会誌 = Journal of Robotics Society of Japan (ISSN:02891824)

巻号頁・発行日

vol.23, no.6, pp.743-751, 2005-09-15

被引用文献数

12

---

Robot audition is a critical technology in creating an intelligent robot operating in daily environments. To realize such a robot audition system, we have designed a missing feature theory based interface between sound source separation and automatic speech recognition (ASR) . In this interface, features distorted by speech separation are detected from input speech as missing features. The detected missing features are masked on recognition to avoid severe deterioration of recognition performance. By using the interface, we developed the robot audition system which recognizes multiple simultaneous speech. We also assess its general applicability by implementing it on three different humanoids, i.e., Honda ASIMO, SIG2, and Replie of Kyoto University. By using three simultaneous speeches as benchmarks, its general applicability was confirmed. When triphone is used and a size of vocabulary is 200 words, the average word correct of three simultaneous speech are 79.7%, 78.7%, and 82.7% for ASIMO, SIG2, and Replie, respectively.

著者

辻野 広司 ケルナー エドガー 桝谷 知彦

出版者

社団法人人工知能学会

雑誌

人工知能学会誌 (ISSN:09128085)

巻号頁・発行日

vol.12, no.3, pp.440-447, 1997-05-01

被引用文献数

6

---

We propose a multi-agent system for hypothetical reasoning based on a large-scale computational theory on essential characteristics of neocortical processing. For a problem-solving on a real world environment, we require both a large-scale computational theory and a robust local computational theory. As a large-scale computational theory, we develop a hypothetical reasoning system by introducing a knowledge-based control on agents and a local commu-nication among agents. These agents communicate each other to reach a globally consistent solution while they locally perform hypothesis generation, representation and evaluation based on a memory-based reasoning as a robust local computational theory. This memory-based reasoning is defined by a principal component analysis, and applies both a deductive reasoning and an inductive reasoning with a least amount of memory that are requisites for hypothetical reasoning. By its multiple representation of same-type knowledge, and its intrinsic local control for decision-state-dependent recall of that knowledge, the proposed agents also serve as symbolic representations of the signal description of a respective feature. Since vision is a typical case for problem-solving by hypothetical reasoning, the proposed general architecture has been used to implement a model on face recognition to verify its performance.

著者

神田 直之 駒谷 和範 中野 幹生 中臺 一博 辻野 広司 尾形 哲也 奥乃 博

出版者

一般社団法人情報処理学会

雑誌

情報処理学会研究報告音声言語情報処理(SLP) (ISSN:09196072)

巻号頁・発行日

vol.2006, no.12, pp.55-60, 2006-02-04

被引用文献数

4

---

複数のドメインを扱う音声対話システムにおいて,対話の文脈や進行に関する特徴量を導入してより精度よくドメイン選択を行う手法を開発したので報告する.本稿ではドメイン選択問題を,応答すべきドメインが,(I)ひとつ前の応答を行ったドメイン,(II)音声認識結果に対する最尤のドメイン,(III)それ以外のいずれかのドメイン,のどれに該当するかを判別する問題と捉える.ドメイン選択の正解を与えた対話データから,対話の文脈や進行に関する特徴量を用いて上記を判別する決定木を学習することにより,ドメイン選択器を構成した.5ドメインのマルチドメイン音声対話システムを用いた10名の被験者による評価実験の結果,音声認識尤度に基づく従来のドメイン選択手法に比べ,ドメイン選択誤りが11.6%削減された.We have developed a robust domain selection method using dialogue history in multi-domain spoken dialogue systems. We define domain selection as classifying problem among (I) the domain in the previous turn, (II) the domain in which N-best speech recognition results can be accepted with the highest recognition score, (III) other domains. We constructed a classifier by decision tree learning with dialogue corpus. The experimental result using 10 subjects shows that our method could reduced 11.6% domain selection error, compared with a conventional method using speech recognition likelihoods only.