In this study, we investigated the relationship between turn-taking and prosody. We considered that to interact smoothly in real-time communication, speakers must show presignals to turn-taking as prosodic features before turn edges. We attempted to discriminate the turn change by the decision tree method using only prosodic features in turn-final accentual phrases that include earlier positions compared with turn-final mora. In the discrimination experiment, we used the corpus of Japanese spontaneous dialogue, and defined prosodic parameters such as F0 contour, power contour and duration. We compared the two parameter conditions for using parameters with and without the final mora of turns. From the results, the accuracy under the conditions of not using the parameters of the final mora is 80%, which is not significantly worse than the result of 83% when using all parameters. Taking into account only prosody was used, we consider this result to be fairly good.
We analyzed the expressivity of recombinant proteins by using data mining methods. The expression technique of recombinant protein is a key step towards elucidating the functions of genes discovered through genomic sequence projects. We have studied the productive efficiency of recombinant proteins in fission yeast, Schizosaccharomyces pombe (S.pombe), by mining the expression results. We gathered 57 proteins whose expression levels were known roughly in the host. Correlation analysis, principal component analysis and decision tree analysis were applied to these expression data. Analysis featuring codon usage and amino acid composition clarified that the amino acid composition affected to the expression levels of a recombinant protein strongly than the effect of codon usage. Furthermore, analysis of amino acid composition showed that protein solubility and the metabolism cost of amino acids correlated with a protein expressivity. Codon usage was often interesting in the field of recombinant expressions. However, our analysis found the weak correlation codon features with expressivities. These results indicated that ready-made indices of codon bias were irrelevant ones for modeling the expressivities of recombinant proteins. Our data driven approach was an easy and powerful method to improve recombinant protein expression, and this approach should be concentrated attention with the huge amount of expression data accumulating through the post-genome era.
A fuzzy constraint satisfaction problem (FCSP) is an extension of the classical CSP, a powerful tool for modeling various problems based on constraints among variables. Basically, the algorithms for solving CSPs are classified into two categories: the systematic search (complete methods based on search trees) and the local search (approximate methods based on iterative improvement). Both have merits and demerits. Recently, much attention has been paid to hybrid methods for integrating both merits to solve CSPs efficiently, but no such attempt has been made so far for solving FCSPs.
In this paper, we present a hybrid, approximate method for solving FCSPs. The method, called the Spread-Repair-Shrink (SRS) algorithm, combines a systematic search with the Spread-Repair (SR) algorithm, a local search method recently developed by the authors. The SRS algorithm spreads (or expands) and shrinks a set of search trees in order to repair constraints locally until, finally, the satisfaction degree of the worst constraints (which are the roots of the trees) is improved. We empirically show that SRS outperforms the SR algorithm as well as the well-known methods such as Forward Checking and Fuzzy GENET, when the size of the problems is sufficiently large.
In this study, we propose a method to construct a system based on a legacy socio-environmental simulator which enables to design more realistic interaction models in socio-environmetal simulations. First, to provide a computational model suitable for agent interactions, an interaction layer is constructed and connected from outside of a legacy socio-environmental simulator. Next, to configure the agents interacting ability, connection description for controlling the flow of information in the connection area is provided. As a concrete example, we realized an interaction layer by Q which is a scenario description language and connected it to CORMAS, a socio-envirionmental simulator. Finally, we discuss the capability of our method, using the system, in the Fire-Fighter domain.
In this paper, a multiagent simulation framework for design of micro chemical processes is proposed. In order to simulate performance of micro chemical device and process behavior under various operational conditions and micro chemical devices, an agent dynamically acquires simulation models collaborating with a DB-agent which relates to a XML-based knowledge database "MDCOs". The classifications of micro chemical devices and simulation models are carried out. Thus the agents can use every simulation models acquiring from DB through the implemented methods declared in the interfaces, which was defined as functions of micro chemical devices. AHP ( Analytic Hierarchy Process ) is adopted to multi-objective evaluation of simulation models to select the preferable simulation model responding to the various objectives of simulation, which are set by the users.
Development of sophisticated anomaly detection and diagnosis methods for spacecraft is one of the important problems in space system operation. In this study, we propose a diagnosis method for spacecraft using probabilistic reasoning and statistical learning with Dynamic Bayesian Networks (DBNs). In this method, the DBNs are initially created from prior knowledge, then modified or partly re-constructed by statistical learning from operation data, as a result adaptable and in-depth diagnosis is performed by probabilistic reasoning using the DBNs. This method fuses and uses both knowledge and data in a natural way and has the both ability which two polar approaches; knowledge-based and data-driven have. The proposed method was applied to the telemetry data that simulates malfunction of thrusters in rendezvous maneuver of spacecraft, and the effectiveness of the method was confirmed.
We think that psychological interaction is necessary for smooth communication between robots and people. One way to psychologically interact with others is through facial expressions. Facial expressions are very important for communication because they show true emotions and feelings. The ``Ifbot'' robot communicates with people by considering its own ``emotions''. Ifbot has many facial expressions to communicate enjoyment. We developed a method for generating facial expressions based on human subjective judgements mapping Ifbot's facial expressions to its emotions. We first created Ifbot's emotional space to map its facial expressions. We applied a five-layer auto-associative neural network to the space. We then subjectively evaluated the emotional space and created emotional regions based on the results. We generated emotive facial expressions using the emotional regions.
In this paper, we first propose a novel interaction model, CEA (Commands Embedded in Actions). It can explain the way how some existing systems reduce the work-load of their user. We next extend the CEA and build ECEA (Extended CEA) model. The ECEA enables robots to achieve more complicated tasks. On this extension, we employ ACS (Action Coding System) which can describe segmented human acts and clarifies the relationship between user's actions and robot's actions in a task. The ACS utilizes the CEA's strong point which enables a user to send a command to a robot by his/her natural action for the task. The instance of the ECEA led by using the ACS is a temporal extension which has the user keep a final state of a previous his/her action. We apply the temporal extension of the ECEA for a sweeping task. The high-level task, a cooperative task between the user and the robot can be realized. The robot with simple reactive behavior can sweep the region of under an object when the user picks up the object. In addition, we measure user's cognitive loads on the ECEA and a traditional method, DCM (Direct Commanding Method) in the sweeping task, and compare between them. The results show that the ECEA has a lower cognitive load than the DCM significantly.
In this paper, we propose a 3-dimensional self-organizing memory and describe its application to knowledge extraction from natural language. First, the proposed system extracts a relation between words by JUMAN (morpheme analysis system) and KNP (syntax analysis system), and stores it in short-term memory. In the short-term memory, the relations are attenuated with the passage of processing. However, the relations with high frequency of appearance are stored in the long-term memory without attenuation. The relations in the long-term memory are placed to the proposed 3-dimensional self-organizing memory. We used a new learning algorithm called ``Potential Firing'' in the learning phase. In the recall phase, the proposed system recalls relational knowledge from the learned knowledge based on the input sentence. We used a new recall algorithm called ``Waterfall Recall'' in the recall phase. We added a function to respond to questions in natural language with ``yes/no'' in order to confirm the validity of proposed system by evaluating the quantity of correct answers.
Profit Sharing is one of the reinforcement learning methods. An agent, as a learner, selects an action with a state-action value and receives rewards when it reaches a goal state. Then it distributes receiving rewards to state-action values. This paper discusses how to set the initial value of a state-action value.
A distribution function ƒ(x) is called as the reinforcement function. On Profit Sharing, an agent learns a policy by distributing rewards with the reinforcement function. On Markov Decision Processes (MDPs), the reinforcement function ƒ(x) = 1/Lx is useful, and on Partially Observable Markov Decision Processes (POMDPs), ƒ(x) = 1/Lw is useful, where L is the sufficient number of rules at each state, and W is the length of an episode.
If episodes are always long, the value of the reinforcement function is little. So the differences of rule values become little, and the agent learns little by using the roulette selection as an action selection. This problem is called as Learning Speed Problem.
If the value of the reinforcement function for an action is very higher than its state-action value, an agent will not select other action. There is a problem when its action is not a optimal action. This problem is called as Past Experiences Problem.
This paper shows that both Learning Speed Problem and Past Experiences Problem are caused by the bad setting between the initial values of a state-action values and the function values of a reinforcement function. We propose how to set the initial values of a state-action values at each state. The experiment shows that an agent can learn correctly even if the length of episode is large. And shows the effectiveness on both MDPs and POMDPs. Our proposed method focuses on the initialization of state-action values and does not limit reinforcement functions. So it can apply to any reinforcement function.
To have an instructional plan guide the learning process is significant to various teaching styles and an important task in an ITS. Though various approaches have been used to tackle this task, the compelling need is for an ITS to improve on its own the plans established in a dynamic way. We hypothesize that the use of knowledge derived from student categories can significantly support the improvement of plans on the part of the ITS. This means that category knowledge can become effectors of effective plans. We have conceived a Category-based Self-improving Planning Module (CSPM) for an ITS tutor agent that utilizes the knowledge learned from learner categories to support self-improvement. The learning framework of CSPM employs unsupervised machine learning and knowledge acquisition heuristics for learning from experience. We have experimented on the feasibility of CSPM using recorded teaching scenarios.
In this paper, we address a solution to density classification tasks using knowledge-based genetic algorithms. Cellular automata (CAs) are used as models of self -organization and emergent computation, and known to have capacity to solve complex problems. It is, however, very difficult to design transition rules that respond to the user's requests, and it prevents the practical application of CAs. Therefore automatic generation of transition rules is studied. We propose a new method to obtain transition rules using knowledge-based genetic algorithms. The knowledge here is a candidate partial solution of the final solution. As a result of infection, the genes of a partial solution are substituted for those of an individual. The purpose of this study is to obtain rules faster than traditional methods. We use the majority decision rule for the knowledge. Experimental results for density classification tasks prove that the proposed method is faster than a conventional method. In addition, the evidence is given that the best transition rules emerge by the partial evolution of the majority decision rule.
We introduce a new convolution kernel for labeled ordered trees with arbitrary subgraph features, and an efficient algorithm for computing the kernel with the same time complexity as that of the parse tree kernel. The proposed kernel is extended to allow mutations of labels and structures without increasing the order of computation time. Moreover, as a limit of generalization of the tree kernels, we show a hardness result in computing kernels for unordered rooted labeled trees with arbitrary subgraph features.
Scale-free and small-world networks receive much attention recently, that are revealed to exist in many natural and artificial systems. There have been several studies on how such networks emerge. In this paper, we propose a novel approach to explain the emergence of different network structures through multi-agent network simulation. Each agent, which represents a node, has rationality and edges to be added are chosen based on mutual common consent. An agent tries to increase its own centrality in the network, and it votes so that its centrality is maximized. Depending on the types of centrality measures, different types of network structures are obtained. This model of network evolution explains emergence of a network where many agents participate in creating it: It includes a social network where each person tries to be more central, and traffic network where each region tries to be more accessible.