畳み込みニューラルネットワーク(CNN:Convolutional Neural Networks)は、人工知能(AI:Artificial Intelligence)に基づく画像識別の分野で広く使用されている。この研究では、CNNによる画像鑑別する新技術の一つResidual Network(ResNet)を用いて大腸ポリープのコンピューター診断支援システム作成について行った。合計で127,610枚の画像(腺腫性ポリープを伴う62,510枚の画像、非腺腫性ポリープ(過形成ポリープなど)30,443枚の画像、および健康な大腸の正常粘膜を有する34,657枚)をAIに学習させてから10-fold cross validationによる12,761枚を使用して検証を行った。ResNetシステムの腺腫ポリープの診断をするための有効性について感度、特異度、PPV、NPV、正診率によって評価した。通常の白色光観察(WLI)において、感度、特異度、PPV、NPV、正診率はそれぞれ98.8%、94.3%、90.5%、87.4%、および92.8%であった。狭帯域光観察(NBI)あるいは色素内視鏡画像(CEI)ではNBI対CEI:感度94.9%対98.2%;特異性93.9%対85.8%;PPV92.5%対81.7%;NPV93.5%対99.9%;正診率91.5%対90.1%であった。ResNetシステムは、大腸ポリープのAI自動診断のため使用できる有力なツールであり臨床現場への導入が期待される。(著者抄録)

BACKGROUND AND AIM: Contrast-enhanced harmonic endoscopic ultrasonography (CH-EUS) is useful for the diagnosis of lesions inside and outside the digestive tract. This study evaluated the value of artificial intelligence (AI) in the diagnosis of gastric submucosal tumors by CH-EUS. METHODS: This retrospective study included 53 patients with gastrointestinal stromal tumors (GISTs) and leiomyomas, all of whom underwent CH-EUS between June 2015 and February 2020. A novel technology, SiamMask, was used to track and trim the lesions in CH-EUS videos. CH-EUS was evaluated by AI using deep learning involving a residual neural network and leave-one-out cross-validation. The diagnostic accuracy of AI in discriminating between GISTs and leiomyomas was assessed and compared with that of blind reading by two expert endosonographers. RESULTS: Of the 53 patients, 42 had GISTs and 11 had leiomyomas. Mean tumor size was 26.4 mm. The consistency rate of the segment range of the tumor image extracted by SiamMask and marked by the endosonographer was 96% with a Dice coefficient. The sensitivity, specificity, and accuracy of AI in diagnosing GIST were 90.5%, 90.9%, and 90.6%, respectively, whereas those of blind reading were 90.5%, 81.8%, and 88.7%, respectively (P = 0.683). The κ coefficient between the two reviewers was 0.713. CONCLUSIONS: The diagnostic ability of CH-EUS results evaluated by AI to distinguish between GISTs and leiomyomas was comparable with that of blind reading by expert endosonographers.

Convolutional neural networks (CNNs) are widely used for artificial intelligence (AI)-based image classification. Residual network (ResNet) is a new technology that facilitates the accuracy of image classification by CNN-based AI. In this study, we developed a novel AI model combined with ResNet to diagnose colorectal polyps. In total, 127,610 images consisting of 62,510 images with adenomatous polyps, 30,443 with non-adenomatous hyperplastic polyps, and 34,657 with healthy colorectal normal mucosa were subjected to deep learning after annotation. Each validation process was performed using 12,761 stored images of colorectal polyps by a 10-fold cross validation. The efficacy of the ResNet system was evaluated by sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and diagnostic accuracy. The sensitivity, specificity, PPV, NPV, and diagnostic accuracy for adenomatous polyps at WLIs were 98.8%, 94.3%, 90.5%, 87.4%, and 92.8%, respectively. Similar results were obtained for adenomatous polyps at narrow-band imagings (NBIs) and chromoendoscopy images (CEIs) (NBIs vs. CEIs: sensitivity, 94.9% vs. 98.2%; specificity, 93.9% vs. 85.8%; PPV, 92.5% vs. 81.7%; NPV, 93.5% vs. 99.9%; and overall accuracy, 91.5% vs. 90.1%). The ResNet model is a powerful tool that can be used for AI-based accurate diagnosis of colorectal polyps.

The use of the Deep Boltzmann Machine for Neuroevolution in the case of Mario AI is discussed in this paper. The scene information in the Mario AI is transformed into a feature space generated by using the Deep Boltzmann Machine. Experimental results show that the proposed method, i.e., Neuroevolution with the Deep Boltzmann Machine outperforms the one with shallow Boltzmann Machine and other methods.

Deep Learning has attracted much attention recently since it can extract features taking account into the high-order knowledge. In this paper, we examine the Deep Boltzmann Machines for scene information of the Mario AI Championship. That is, the proposed method is composed of two parts: the DBM and a recurrent neural network. The DBM extracts features behind perceptual scene information, and it learns off-line. On the other hand, the recurrent neural network utilizes features to decide actions of the Mario AI agents, and it learns on-line by using Particle Swarm Optimization. Experimental results show the effectiveness of the proposed method.

This article has no abstract.

In this paper, we propose a system to collect human behavior in detail with higher-level tags such as attitude, accompanying-person, expenditure, tasks and so on. The system makes it possible to construct database to model higher-level human behaviors that are utilized in context-aware services. The system has two input methods i.e., on-the-fly by a smartphone and post-processing by a PC browser. On the PC, users can interactively know where and when they were, which results in easily and accurately constructing human behavior database. To decrease numbers of operations, users' repetitions and similarity of among users are used. According to experiment results, the system successfully decreases operation time, higher-level tags are properly collected and behavior tendencies of the users are clearly captured. © 2012 Crown Copyright.

We have shown that the Isomap, one of the most famous Manifold Learning method, is suitable for Neuroevolution of mobile robots with redundant inputs. In the proposed method, a large number of high dimensional inputs are collected in advance. The Manifold Learning method yields the low dimensional space. Evolutionary Learning is carried out with the low dimensional inputs, instead of the original high dimensional inputs. In this paper, the Isomap and Manifold Sculpting are compared by using Mario AI Championship.

Recently, we suggested that soft computing techniques such as NNs, MOEA-II, and FEP could be utilized for making educational games much more exciting. In this paper, we present game playing results of the original COMMONS GAME and two kinds of the modified COMMONS GAME having been carried out in several universities in Kansai area in Japan. These gaming results suggest that the modified COMMONS GAMES utilizing soft computing techniques such as NNs, MOEA-II, and FEP is by far the best in order to be utilized for letting people consider seriously about the "COMMONS" issue which is quite important not only for human beings but also all of the creatures in this planet.

Mario AI is one of competitions on Computational Intelligence. In the case of video games, agents have to cope with a large number of input information in order to decide their actions at every time step. We have proposed the use of Isomap, a famous Manifold Learning, to reduce the dimensionality of inputs. Especially, we have applied it into scene information. In this paper, we newly extend to enemy information, where the number of enemies is not fixed. Hence, we introduce the proximity metrics in terms of enemies. The generated low-dimensional data is used for input values of Neural Networks. That is, at every time step, transferred data by using a map from raw inputs into the low-dimensional data are presented to Neural Networks. Experimental results on Mario AI environment show the effectiveness of the proposed approach.

In this paper, we suggest that soft computing techniques such as NNs, MOEA-II, and FEP could be utilized for making the original COMMONS GAME much more exciting. Several game playing results by our students confirm the effectiveness of our approach.

The optimization of the number and the alignment of sensors is quite important task for designing intelligent agents/robotics. Even though we could use excellent learning algorithms, it will not work well if the alignment of sensors is wrong or the number of sensors is not enough. In addition, if a large number of sensors are available, it will cause the delay of learning. In this paper, we propose the use of Manifold Learning for Evolutionary Learning with redundant sensory inputs in order to avoid the difficulty of designing the allocation of sensors. The proposed method is composed of two stages: The first stage is to generate a mapping from higher dimensional sensory inputs to lower dimensional space, by using Manifold Learning. The second stage is using Evolutionary Learning to learn control scheme. The input data for Evolutionary Learning is generated by translating sensory inputs into lower dimensional data by using the mapping.

The accessibility of information on web-space is considered to be focused on the characteristics of hyperlinks. Although the reasons to put a hyperlink are multifarious, only one kind of link exists in HTML structure. As a result of this, users have to search for links by using syntactically techniques so as to retrieve information in web-space. In this paper, four kinds of hyper links are introduced into HTML, in order to improve the accessibility of information in web-space: Outbound links, Internal links, Navigation links and Intra-unit links. Designers and users can share the image of the link target page by using these kinds of links. Furthermore, the information flow between designers and users is described using mathematical framework, and experiments are carried out in order to verify the scheme.

EDA-RL, Estimation of Distribution Algorithms for Reinforcement Learning Problems, have been proposed by us recently. The EDA-RL can improve policies by EDA scheme: First, select better episodes. Secondly, estimate probabilistic models, i.e., policies, and finally, interact with the environment for generating new episodes. In this paper, the EDA-RL is extended for Multi-Objective Reinforcement Learning Problems, where reward is given by several criteria. By incorporating the notions in Evolutionary Multi-Objective Optimization, the proposed method is enable to acquire various strategies by a single run.

We previously proposed evolutionary fuzzy systems of playing Ms.PacMan for the competitions. As a consequence of the evolution, reflective action rules such that PacMan tries to eat pills effectively until ghosts come close to PacMan are acquired. Such rules works well. However, sometimes it is too reflective so that PacMan go toward ghosts by herself in longer corridors. In this paper, a critical situation learning module is combined with the evolved fuzzy systems, i.e., reflective action module. The critical situation learning module is composed of Q-learning with CMAC. Location information of surrounding ghosts and the existence of power-pills are given to PacMan as state. This module punishes if PacMan is caught by ghosts. Therefore, this module learning which pairs of (state, action) cause her death. By using learnt Q-value, PacMan tries to survive much longer. Experimental results on Ms.PacMan elucidate the proposed method is promising since it can capture critical situations well. However, as a consequence of the large amount of memory required by CMAC, real time responses tend to be lost.

We have suggested that utilization of soft computing techniques such as GAs and EAs could contribute a lot for making the original COMMONS GAME much more exciting. In this paper, we try to find an answer concerning to the question "Which game is the best for letting people consider seriously about the commons among the three games (the original COMMONS GAME, the modified COMMONS GAME utilizing GAs & NNs, and the modified COMMONS GAME utilizing EAs & NNs) ?". Several game playing by our Students confirm that the modified COMMONS GAME utilizing EAs & NNs can provide the best chance for letting players consider seriously about the commons.

In this paper, we suggest that Evolutionary Algorithms could be utilized in order to let the COMMONS GAME, one of the most popular environmental games, become much more exciting. In order to attain this objective, we utilize Multi-Objective Evolutionary Algorithms to generate various skilled players. Further, we suggest that Evolutionary Programming could be utilized to find out an appropriate point of each card at the COMMONS GAME. Several game playings utilizing the new rule of the COMMONS GAME confirm the effectiveness of our approach. © 2007 Springer-Verlag Berlin Heidelberg.

In this paper, we focus on game rule design by using two Evolutionary Computations. The first EC is a Multi-Objective Evolutionary Algorithm in order to generate various skilled players. By using acquired skilled players, i.e., Pareto individuals in MOEA, another EC (Evolutionary Programming) adjusts game rule parameters he, an appropriate point of each card in the COMMONS GAME.

In this paper, we suggest that EAs could be utilized for making the Environmental Game become much more exciting. In particular, in order attain this objective, we try to utilize Evolutionary Algorithms in the following steps: (1) First, we consider a new way for producing a return for each player. (2) Second, we utilize Multi-Objective Evolutionary Algorithms (MOEAs) to generate various skilled players whose decision making concerning the investment to the chemical goods (and consideration toward environmental protection) can strongly affect the environmental state of the sea. (3) Further, we utilize Evolutionary Programming (EP) to derive appropriate rules which could be used to help players fully enjoy game playing. Several game playing (utilizing the new rule) by our students suggest the effectiveness of our approach.

The accessibility of information on web-space is considered to be focused on the properties of hyperlinks. Although the reasons to put in a hyperlink are multifarious, only one kind of link exists in HTML structure. As a result of this, users have to search for links by using syntactically techniques so as to retrieve information in web-space. In this paper, four kinds of hyper links are introduced into HTML, in order to improve the accessibility of information in web-space: External links, Internal links, Navigation links and Intra-unit links. Designers and users can share the image of the link target page by using these kinds of links.

Estimation of Bayesian Network Algorithms which adopt Bayesian Networks as the probabilistic model were one of the most sophisticated algorithms in the Estimation of Distribution Algorithms. However the estimation of Bayesian Network is key topic of this algorithm, conventional EBNAs adopt greedy. searches to search for better network structures. In this paper, we propose a new EBNA which adopts Genetic Algorithm to search the structure of Bayesian Network. In order to reduce the computational complexity of estimating better network structures, we elaborates the fitness function of the GA module, based upon the synchronicity of specific pattern in the selected individuals. Several computational simulations on multidimensional knapsack problems show us the effectiveness of the proposed method.

In this paper, a method for measuring the shape of a columnar object with specular surfaces by a slit ray projection method is proposed. Although the slit ray projection method is effective for measuring the shape of an object having diffuse reflective characteristics, applying the method to an object with specular surfaces has hitherto been difficult. In this paper, a triangulation equation for a columnar object with specular surfaces is derived and used as the basis of a method for measuring the shape of an object having specular surfaces using applicable slit rays. The basic principle is that the angle of incidence of a slit ray reflected from the measured object into an image sensor is restricted specifically by the special design of the optical system. In addition to the theoretical study of the proposed principle, a system for measuring the shape of a columnar specular object and a diffuse object was created and its effectiveness was verified. © 2002 Wiley Periodicals, Inc. Syst. Comp. Jpn.

This paper applies our state construction method by ART Neural Network to Robot Navigation Problems. Agents in this paper consist of ART Neural Network and Contradiction Resolution Mechanism. The ART Neural Network serves as a mean of state recognition which maps stimulus inputs to a certain state and state construction which creates a new state when a current stimulus input cannot be categorized into any known states. On the other hand, the Contradiction Resolution Mechanism (CRM) uses agents' state transition table to detect inconsistency among constructed states. In the proposed method, two kinds of inconsistency for the CRM are introduced: "Different results caused by the same states and the same actions" and "Contradiction due to ambiguous states." The simulation results on the robot navigation problems confirm us the effectiveness of the proposed method.

In this paper, we propose a constitution method of game player agent that adopts a neural network as a state evaluation function for the game player, and evolves its weights and structure by Evolutionary Strategy (ES). In this method, we attempt to acquire better state evaluation function by evolving weights and structure simultaneously.

This paper applies our state construction method by ART Neural Network to Robot Navigation Problems. Agents in this paper consist of ART Neural Network and Contradiction Resolution Mechanism. The ART Neural Network serves as a mean of state recognition which maps stimulus inputs to a certain state and State construction which creates a new state when a current stimulus input cannot be categorized into any known states. On the other hand, the Contradiction Resolution Mechanism (CRM) uses agents’ state transition table to detect inconsistency among constructed states. In the proposed method, two kinds of inconsistency for the CRM are introduced: “Different results caused by the same states and the same actions" and “Contradiction due to ambiguous states.” The simulation results on the robot navigation problems confirm us the effectiveness of the proposed method.

Recently, many researchers have studied for applying Fuzzy Classifier System (FCS) to control mobile robots, since the FCS can easily treat continuous inputs, such like sensors and images by using fuzzy number. By using the FCS, however, only reflective rules are acquired. Thus, in proposed approach, an additional Genetic Algorithm in order to search for strategic knowledge, i.e., the sequence of effective activated rules in the FCS, is incorporated. That is, proposed method consists of two modules: an ordinal FCS and the Genetic Algorithm. Computational experiments based on WEBOTS, one of Khepera robots' simulators, are confirmed us the effectiveness of the proposed method.

In this paper, we will introduce a new Coevolutionary Genetic Algorithm with schema extraction by machine learning techniques. Our CGA consists of two GA populations: the first GA (H-GA) searches for the solutions in the given problems and the second GA (P-GA) searches for effective schemata of the H-GA. We aim to improve the search ability of our CGA by extracting more efficiently useful schemata from the H-GA population, and then incorporating those extracted schemata in natural manner into the P-GA. Several computational simulations on multidimensional knapsack problems confirm us the effectiveness of the proposed method.

In game programming for Chess, Reversi, and so on, it is difficult to evaluate each state and to decide which attributes should be used for state evaluation. Hence, state evaluation functions are often designed by using heuristics of human. In this paper. we adopt a neural network as a state evaluation function for game player, and evolve its weights and structure by Evolutionary Strategy (ES). In the proposed method, we attempt to acquire better state evaluation function by evolving weights and structure simultaneously.

We discuss on adaptability of Evolutionary Computations in dynamic environments. Hence, we introduce two classes of dynamic environments which are utilizing the notion of Constraint Satisfaction Problems: changeover and gradation. The changeover environment is a problem class which consists of a sequence Of the constraint networks with the same nature. On the other hand, the gradation environment is a problem class which consists of a sequence of the constraint networks such that the sequence is associated to two constraint networks, i.e., initial and target., and all constraint networks in the sequence metamorphosis from the initial constraint network to the target constraint network. We compare Coevolutionary Genetic Algorithms With SGA in computational simulations. Experimental results On above dynamic environments confirm us the effectiveness Of Our approach, i.e., Coevolutionary Genetic Algorithm.

In this paper, we propose a new incremental state segmentation method by utilizing information of agents state transition table which consists of tuple of (state, action. state) in order to reduce the effort of designers and which is generated by ART Neural Network. In the proposed method, if inconsistent situation in the state transition table is observed, agents refine their map from perceptual inputs to states such that such inconsistency is resolved. We introduce two kinds of inconsistency.. i.e., "Different Results Caused by the Same States and the Same Actions" and "Contradiction due to Ambiguous States." Several computational simulations on cart-pole problems confirm us the effectiveness of the proposed method.

In this paper, we discuss how many satisfiable solutions Genetic Algorithms can find in the problem instance of Constraint Satisfaction Problems at single execution. Hence, we propose a framework of a new fitness function which can apply to traditional fitness functions. However the mechanism of proposed fintess function is quite simple, several experimental results on a variety of instances of General Constraint Satisfaction Problems demonstrate the effectiveness of proposed fitness function.

In this paper, we discuss the adaptability of Coevolutionary Genetic Algorithms on dynamic environments. Our CGA consists of two populations: solution-level one and schema-level one. The solution-level population searches for the good solution in a given problem. The schema-level population searches for the good schemata in the former population. Our CGA performs effectively by exchanging genetic information between these populations. Also, we define Dynamic Constraint Satisfaction Problems as such dynamic environments. General CSPs are defined by two stochastic parameters: density and tightness, then, Dynamic CSPs are defined as a sequence of static constraint networks of General CSPs. Computational results on DCSPs confirm us the effectiveness of our approach.

A new genetic algorithm (GA) which coevolves individuals and schemata is presented. The proposed method is inspired by the idea that effective utilization of symbolized information on solution space may be useful for GA-based search for solutions. This symbolized information may be represented as schemata, and schemata can be expressed in subspaces which will have high average fitness values, i.e. building blocks. In this paper, a specialized GA model for schemata search is introduced. This GA model directory tries to search for useful schemata that have not been yet discovered by the search in GA model. Thus, the GA for schemata search is expected to guide the traditional GA-based search. Several analyses based on various computer simulations reveal the effectiveness of the proposed method.

本書では，ニューラルネットをはじめとするファジー工学で代表されるソフトコンピューティングの基礎と応用を学ぶ。具体的な応用事例として生体信号処理とパターン認識，株価予測，コンピュータゲーミングについても詳解している。さらに遺伝的アルゴリズムや進化戦略などの進化計算についても概説し，適用範囲の広さを学べる内容となっている。さらにはファジー集合・推論・制御の基礎とその応用について概説している。また各章末には演習問題が用意されており，本書1冊でソフトコンピューティングの概要が初学者でもわかるように工夫されている。

本研究課題では、グラフの探索を進化計算にて行うものである。先行研究のEDA-GKでは、特徴空間で問題を捉えることにより、遺伝子型から表現型への写像の凸凹さを解消できていると考えられる。ところが、研究を押し進めていく上で、先行研究のEDA-GKでは，カーネル関数が対象としている特徴空間によって，問題によっては機能しないことが判ってきた。例えば、前ページの最小パス距離カーネルの場合、解が全結合に近いグラフであった場合任意のノード間の距離がほとんど1となり、最適解付近のグラフのヒストグラムに違いがなくなる。結果、ランダムなエッジの追加・削除によるランダム探索に近い挙動を示すことになる。そこで、本研究課題では、Deep Graph Kernelの利用を検討する。Deep Graph Kernelはグラフカーネルが対象とする特徴に対して、それらの関係性も学習する ものである。word2vecにあるような単語の埋め込みをグラフに拡張したものである。特徴空間を構成するときに、特徴間の関係を埋め込むことより、カーネル関 数の特性に左右されない安定した探索を実現する。事前学習が必要となるが、領域知識が提供できる問題クラスなら適用可能であると考え、特に、提案手法を有機薄膜太陽光電池で評価することを考える。 有機薄膜太陽電池は、比較的大きめの分子構造となるため、スクラッチで原子を組み合わせても適切な解表現を探索することができない。そこで、既存の有機薄膜太陽光電池に用いられている分子構造の部分構造をモジュールとして、モジュールの組み合わせを探索するようにした。さらに，QDFという深層学習の一種を用いることにより、より大規模な分子構造をもつ解に対して、進化計算が適用できるようになった。

The distribution of estimation algorithm EDA-GK using graph kernels is extended in this study. Conventionally, it has been difficult to achieve good performance in evolutionary algorithms using graphs as individuals because the mapping from genotype to phenotype is rugged. In this study, we aim to extend the application area of the algorithm. A mixed kernel is constructed and applied to graph identification problems with scale-free and small-world properties. Simultaneously, it is applied to the order/degree problem and the eigenvalue maximization problem, which are coped with in the field of graph theory.

This research project extends the estimation of distribution algorithms to cope with graph individuals. The proposed method employ graph kernels to estimate the distribution of graph individuals. That is, the proposed method search for better solutions in the feature space. The results can be summarized as follows: The implementations of algorithms is re-designed such that the sampling method is examined. Moreover, the proposed method extended to cope with nodes with types. Moreover, the proposed method is applied to the Order/Degree problems, which is examined in graph study groups. The experimental results showed that the proposed method is competent with other conventional method designed for solving the Order/Degree problems, meanwhile, the proposed method is general-purpose algorithms.

The project aims to develop a new systems theory based on a viewpoint where further benefits of inconveniences are appreciated. Analyzing case examples of such design field as mechanical, mechanism, industrial and interface design results in many papers and some books (e.g. DOJIN sensyo no.42, BUNSYU sinsyo no.901). Preliminary reports of our research can be found in our web pages (ruben-eki.jp, www.facebook.com/fuben.eki). Practical applications of our research have risen to national prominence. An application to article design has been reported by many media, and an application to mechanism design, called BiblioBattle, which was incorporated in a plan decided by the Cabinet, has grown to become a nationwide convention.

This research project has proposed two sorts of novel Estimation of Distribution Algorithms (EDAs). The first EDA is the Estimation of Distribution ALgorithms for solving Reinforcement Learning Algorithms, called EDA-RL. For this algorithm, we have investigated the detection of slip states, and the estimation of the structure of the probabilistic model in the EDA-RL. Another proposed EDA is the EDA for graph-related problems. This algorithm can cope with individuals represented by graphs. By using graph kernels, we do not have to take care of ruggede fitness landscpaes caused by individual representations. During the project peirod, we have published journal articles and international coneference papers. Moreover, we have two oppotunity of the invited talks including a talk in Taiwan.

The ambient environment that we humans face in reality is one in which almost all is changed with time characteristics. Therefore, in order to establish the utility of soft computing techniques, it is necessary to confirm the validity of the soft computing techniques under so-called non-stationary environment whose characteristics changed with time. In this study we propose the fusion system of soft computing technology (such as NNs) and learning automata and check the effectiveness of the fusion system through the various applications such as stock price prediction, computer gaming and etc.

This study proposes a knowledge acquisition method by cooperative distributed evolutionary algorithms with divide and conquer approach. By decomposing problems spaces in advance, we can expect that the proposed method enables evolutionary algorithms to maintain the diversity in genetic information. We employ Estimation of Distribution Algorithms as Evolutionary Algorithms. We have concentrated to extend the EDAs to acquire knowledge during the research period.

今年度までに展開してきた研究を総括し、「情報土壌学」という新しい学域の基本的成り立ちの骨格を与えるとともに、さまざまな領域においてその展開の具体的な形を示してきた。 まず、予め設定(想定)された命題群の真・偽や成立・不成立を判定するという二分法をベースとして、想定内での状況の絞り込みの手がかりとしての情報の在り方、すなわち、シャノン流の情報概念とは異なり、第二の情報の在り方として、想定外の事象生起に対して新たな包括的な視点を導入しつつ諸事象包摂し、それらに時空の秩序を与えるような働きをする情報概念を提案した。前者が部分や要素への「分化・排除」を指向する働きであるのに対し、後者は全体への「統合・包摂」を指向するものであり、互いに相補的な情報の働きということができる。 情報土壌学では、主として後者のような情報を扱う枠組みとして、記号論・論理学の始祖の一人とされるパースによる述語論理式の表現である存在グラフに加えて、離散事象システムの事象推移表現であるペトリネット(発生ネット)を併合した新たな数学的表現形式を導入して、この包摂的情報の働きを表現し、解析する枠組み構築した。これは、ライプニッツの提唱した時空概念に即して情報群と人間集団や社会との連関構造を捉えるものであり、社会秩序や風土の成り立ちと包摂的な情報の働きの接点を解明する手掛かりを成すものである。その上に立って、倫理の問題やケアの世界における共感的・共存在的・即興的なコミュニケーションの在り方をライプニッツ時空の共生成という立場から明らかにするとともに、人工物の創造的なデザインやその使用における人工物との関係性やインタラクションの働き、とりわけ不便さや制約の効用などについて新たな解明と提案を行った。また、包摂的な情報と社会集団の形成のようなグループダイナミックスの連関の深層構造についての解明を行い、この情報土壌学の視点の有用性を明らかにした。

手間いらずで効率的に要求が満たせる「便利な道具や方式」よりも、むしろ不便な道具や方式に、能動的工夫の余地・対象系の物理的理解促進・自己肯定感の醸成、などの効果がある。我々はこれらを積極的に評価する「不便益」という考え方を提唱し、この視点からの新たなシステム設計方法論の構築を試みた。最終年度にあたる2008年度に得られた成果を以下にまとめる。 [不便益の総論:]初年度から継続する各種講演やOSを年に数回開催し、そこで得られた多くの知見の整理を通して不便益の輪郭を明らかにして、HI学会論文誌に総説論文としてまとめた。 [不便益の各論:]システムデザインにおける「便利」を基礎づけるものとして、因果に基づく明瞭な説明・分類/分割による高いモジュラリティ・あいまいさの無い推論・中央集権的システム構成に注目し、それらが援用できない「不便」、すなわち均衡(バランス)に基づく説明・明瞭な分割ができないこと・解釈(推察)の多様性・分権的制御などから得られる益を積極的に活用する考え方を提出するとともに、それらのいくつかには具体的な活用方法や数理的基盤を与えた。 [不便益の応用:]不便益に関する知見をシステムデザインに応用することによって、その有効性を検証した。適用対象としては、情報伝達に関して「見るのではなく触れることしかできない絵画」や音と画像による情報伝達におけるテロップの効用分析、解釈の多様性に関して比喩表現やピクトグラムによる非言語コミュニケーションの有効性検証と、エージェントシミュレーション、デザインの実践として月に一度のインクルーシブデザインワークショップ(塩瀬)、などを実施した。

Recently, there has been a growing trend to promote recycling society in Japan. This study therefore aim to realize an automatic separating system of plastic bottles for reuse. In 2005, we tried to construct to the pro type system to measure the shape and to identify the material of the transparent object, which consist of laser light and ultrasonic wave. The transparent object has the primitive shapes, which are a cylinder, a rectangle prism, a square pyramid, and a sphere. First, we designed and built both the laser rangefinder system, which consist of the 6-axis robot and the imaging system by our own making and can measure the transparent object such as plastic bottle. From experimental results, it is clear that the prototype laser rangefinder is effective to measure the shape of the transparent objects. Next, we made the prototype measurement system using the ultrasonic sensor array, which can measure the shape and identify the material of the transparent object. The prototype measurement system has allow to identify the shape of the primitive objects. In 2006, we made the combined system of laser rangefinder and the ultrasonic measurement system. We verified the usefulness of the combined system by experiment. The measured object is commercially available plastic bottles. From experimental results show the effectiveness of the measurement and the identification of the area of interest. In concluding, the research attained the originally planned object.

To address the situation where recent methodologies of Universal Design (UD) become estranged from "the principles of UD" originated with R.Mace, we examined guidelines for UD by employing notions of systems science. The guidelines regard re-designs by alterations of physical quantities as a last resort. The guidelines aim to focus designers' intentions to qualitative gropes. The axis of our research is our assumption of a sufficient condition on UD : "instead of enforcing an unique optimized operation-sequence, they are open to various interpretations" and "generous constraints afford possible operations" and "interactions between a operation and the following alterations of affordance can be interpreted as a conversation." We broke this axis down into three sub-axes that are interdependent with each other. The first sub-axis was to elucidate the concepts : "generous constraint" and "interaction as conversation." For this purpose, we proposed models of "generous constraint" by modal logics and by the arbitrariness of information channels in terms of Qualitative Information Theory. We also proposed a framework of a design support system based on "interaction as conversation" defined from the viewpoint of alethic, deontic and temporal modalities, and teleological-causal necessities. The second sub-axis was bridging from qualitative to quantitative design processes. Adopting genetic algorithms, we developed a constraint solver that treats qualitative statements as constraints then decides quantitative parameters. The third sub-axis was actual proofs. The results of first and second axes were applied to several designs. For example, we participated in the project of designing rest rooms in the ward of a rehabilitation center in Osaka pref., that of designing entrance gates of unattended bicycle parking lots with a venture company in Kyoto city, and that of web pages. Furthermore, for comparing with UD, we invited Ms. Jullia Cassim, the leading expert of Inclusive Design, and carried out a workshop.

近年,ニューラルネットや強化学習,進化計算のような学習機構を取り入れた自律移動ロボットに関する研究が盛んである.しかしながら,現状では,たとえ高性能な学習機構を取り入れていても,地面のすべりや背景といった環境特有の情報,エージェントのアクチュエータやセンサの特性といった生態学的ニッチ固有の制約構造を巧く活用しないとロボットは全く機能しない.本研究では,自律的に知覚・行動を行う強化学習エージェントに対して与えられた生態学的ニッチ固有の制約構造を活用するための状態分割法を提案しその有用性を示した.提案手法では,知覚-行動-知覚という系列について,状態分割の整合性を検査し,矛盾が検出されるたびにさらなる分割を生じることによりその矛盾解消を試みる.すなわち,ある時刻における知覚入力に対して状態を割り当てるような状態分割器を考える.そして,行動を行った結果として生じる知覚入力を状態分割器に与え,状態を得る.系列をエージェントが行動するたびに蓄えていき,矛盾検出とその解消を行う.提案手法では、階層的な情報処理による状態分割を行うメカニズムを有し、階層は情報の縮約を、上位層は階層で縮約を行ったニューロンが発火した際に、次時刻ではどのような知覚変化を引き起こすかを予測している。この予測の精度にもとづいて状態分割を構成している。本年度は、この研究課題で新たに提案を行った上記の状態分割法について、車の山登り問題へ適用した研究について2004年7月にブダペストで行われたIJCNN04にて成果発表を行った。また状態だけではなく行動の分節化を行うようなアルゴリズムの拡張について検討した。

In this research, we have developed a new triangulation-based laser rangefinder for measuring the 3-D shapes of objects with hybrid reflectance, which combines Lambertian reflectance and specular reflectance. Ordinary laser rangefinding techniques at present have rarely been applied to the measurement of the shapes of this type of object, since the equations for the computation of 3-D position based on tnangulation do not hold for objects with specular reflectance. Like the ordinary rangefinder, the proposed rangefinder consists of a light-stripe generator, a lens system, and an image sensor. The lens system hes at the core of the scheme of our rangefinder. It consists of a field stop, a lens, and shield masks. The field stop is placed in front of the lens and parallel to the lens, and has horizontal slits. The shield masks are located between the lens and the image sensor, and composed of multiple thin plates which stand vertically at a constant angle with respect to the lens. We measured shape measurements of object with hybrid surfacesThe experimental results demonstrate that the proposed rangefinder allows the measurement of hybrid objects with almost the same level of accuracy as that of Lambertian object, and specular objects Since the rangefinder does not require a reflectance model of the object's surface, it has the advantages of higher accuracy and reliability over other shape extraction techniques for Lambertian surfaces and/or specular surfaces. The proposed rangefinder is especially available for industrial purposes since a consistent accuracy and high reliability are assured even for objects with unknown reflectance properties. In conclusion the rangefinder is promising for many common industrial tasks including measurement of the shapes of metal objects and the inspection of solder joint surfaces.

Many practical tasks in industry often require the determination of the three-dimensional (3-D) shapes of objects with specular surfaces, such as polished metal parts, soldered joints or welded areas. Although several methods exist for the determination of such information, these methods have been only partially succeeded in determining of shape information due to the following two limitations : (1) These methods have generally assumed the objects' reflectance properties to be exclusively Lambertian or specular. (2) The accuracy of shape measurement has strongly deepened on the accuracy of image intensity measurement. In 1998, we have proposed a new laser rangefinder for measuring the 3-D shapes of specular objects. The basic concept of the proposed rangefinder is to limit the incident angle of a light-strip upon the image sensor to a constant angle using shield masks, resulting in the ability to apply a triangulation system to the measurement of specular objects. And, we derived the equations to determine an objects' 3-D positions by triangulation and applied them to the rangefinder. In 1999, we designed and built a prototype rangefinder, which fulfils the requirements of our original idea. Experiments using the rangefinder successfully achieve accurate measurements using light-stripe projection. The rangefinder is promising for many common industrial tasks including measurement of the shapes of polished metal objects and the inspection of solder joint surface. Important factors limiting the rangefinder's practical applications are its field of view, depth of field and speed when applied to specular objects.