著者
木村 元 山村 雅幸 小林 重信
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.11, no.5, pp.761-768, 1996-09-01
被引用文献数
60

Many conventional works in reinforcement learning are limited to Markov decision processes (MDPs). However, real world decision tasks are essentially non-Markovian. In this paper, we consider reinforcement learning in partially observable MDPs(POMDPs) that is a class of non-Markovian decision problems. In POMDPs assumption, the environment is MDP, but an agent has restricted access to state information. Instead, the agent receives observation containing some information about states of the MDP. Also we focus on a learnig algorithm for memory-less stochastic policies that map the immediate observation of the agent into actions: The memory-less approaches are suited for on-line and real-time adaptive systems that have limited memory and computational resources. Then, the following mathematical results are got. First, it can improve its policy to maximize immediate reward by stochastic gradient ascent without estimating any state or immediate reward. Second, it can improve the policy to maximize discounted reward in an initial state by stochastic gradient ascent without estimating any state, immediate reward or discounted reward. The above advantages are remarkably effective in POMDPs, because it is not required to estimate any states, immediate reward or discounted reward explicitly. Making use of these results, we present an incremental policy improvement algorithm to maximize the average reward in POMDPs. We ensure the rational behavior of the proposed algorithm in a simple experiment.
著者
藤本 和則 賀沢 秀人 佐藤 浩史 阿部 明典 松澤 和光
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.15, no.1, pp.61-64, 2000-01-01
被引用文献数
8

Decision Support for Internet Users, Called DSIU, is an area of research for providing decision support for Internet users by using information on the Internet. DSIU aims to provide decision support with logicalexplanation taking account of user's preference. By using information extraction techniques, DSIU handles the names of various substances, e.g., electronic products, persons, and places, and so on, and constructs the explanations in terms of their properties. This paper describes the DSIU particularly form a viewpoint of realizing the DSIU and giving contributions to society in the near future. The information of DSIU is available at http://www.kecl.ntt.co.jo/DSIU/.
著者
古崎 晃司 溝口 理一郎
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.20, no.6, pp.707-714, 2005-11-01
被引用文献数
11
著者
山本 恭裕 高田 眞吾 中小路 久美代
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.14, no.1, pp.82-92, 1999-01-01
被引用文献数
17

The goal of this study is to design and build a computer system to support the basic cognitive activity of "writing" in a more natural and effective manner. The paper starts with a description of a writing process, followed by an overview of existing models on writing. Then, the notion of "Representational Talkback" is proposed as an important aspect in supporting collage-style writing. Representational Talkback is defined as "feedback from externally represented artifacts." The ART (Amplifying Representational Talkback) system is implemented based on this notion, focusing on the role of meta-comments in writing. The goal of the system is twofold : (1) to support collage-style writing of a document, and (2) to observe how people "write" using ART. The paper concludes with a discussion of the result of a study on how people "write" using ART with an eye towards extending the notion to other types of cognitive activities.
著者
松尾 利行 西田 豊明 星本 健一
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.1, pp.68-77, 1997-01-01
被引用文献数
3

In this paper, we describe a practical method of extracting, structuring, summarizing, and integrating technical information from technical papers in metallurgy. The heart of the method is packets of domain specific knowledge called KP (Knowledge Pieces) in which procedures for extracting and structuring technical information from technical papers are embedded. We studied information structure of ten technical papers in metallurgy and constructed about a hundred KPs. We implemented a system called METIS which takes technical papers in metallurgy encoded in a mark-up language and produces a varieties of summaries and surveys including structured technical summary, visual display of similarites and differences of relevant papers, and Cause-effect relations. We have undertaken qualitative and quantitative evaluation of METIS against 106 technical papers so far. The evaluation demonstrates the reliability and robustness of our method.
著者
桐山 孝司 冨山 哲男 吉川 弘之
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.6, no.3, pp.426-434, 1991-05-01
被引用文献数
26

Integration of design object models is one of the expected roles of intelligent CAD systems. This paper deals with maintenance of relationships among models. We examine the nature of models and show that knowledge about relationships among background theories is crucial for the integration. Based on this discussion, we propose the metamodel mechanism, a new framework for integrated design object modeling. The idea of the metamodel mechanism is to utilize a qualitative model in order to represent dependency among concepts of which the models consist. The metamodel is refined through the design process by four operations, viz. instantiation, unification, specialization, and delegation. We also show an implementation of the metamodel mechanism.
著者
佐藤 俊治
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.15, no.6, 2000-11-01

本論文は神経生理学や心理学的知見を導入して, パターンの多様性に影響されない視覚神経回路網モデルの構築を目的としており, 6章より構成される.第1章の「序論」に続いて, 第2章では, 既に提案されている視覚神経回路網モデルであるネオコグニトロンの構成・学習方法について定式化するとともに, 回転したパターンに頑健性がないことを実験により確認した.第3章では, 新しいボトムアップ型神経回路モデル(回転対応型ネオコグニトロン)を提案している.実際に手書き数字を用いた数値実刑により, パターンの変形・位置ずれ・拡大縮小・ノイズだけでなく, パターンの回転にも完全に頑健であることを示した.第4章では, 回転対応型ネオコグニトロンを含むネオコグニトロン型神経回路モデルの学習過程を解析し, その結果から高速に学習を行なうアルゴリズムを提案している.本アルゴリズムを用いることで, 認識性能に影響を及ぼすことなく, 学習時間が約1/680に短縮することを確認している.第5章では, 回転した文字を必ずしも瞬時に認識せず心的回転により初めて認識するというヒトの認識機能を実現する視覚モデルを提案している.数値実験により, パターンの多様性に頑健であることを明らかにした.また, 鏡像回転パターンに対する提案モデルの挙動が心理学的事実と符号するという興味深い結果も得られた.第6章「結論」では本論文の成果をまとめ, 今後の課題を述べている.
著者
アルモアリム フセイン 秋葉 泰弘 金田 重郎
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.421-429, 1997-05-01
被引用文献数
7

This paper studies the problem of learning decision trees when the attributes of the domain are tree-structured. Quinlan suggests a pre-processing approach to this problem. When the size of the hierarchies used is huge, Quinlan's approach is not efficient and effective. We introduce our own approach which handles tree-structured attributes directly without the need for pre-processing. We present experiments on natural and artificial data that suggest that our direct approach leads to better generalization performance than the Quinlan-encoding approach and runs roughly two to four times faster.