Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IOS Press  

In this study, we display preliminary results for harnessing fuzziness of yet-another fuzzy rule-bases. They are based on the pragmatic rule-design (PRD), which has been proposed by the authors. The PRD is novel since a pragmatic rule is not an "IF-THEN" rule nor an artificial neural network, and does not represent a stimulus-response relation. A pragmatic rule is a vector of relative characteristics of effective responses in itself. In the original PRD, the fuzziness in discretizing a system state is too surplus. Restricting such fuzziness may improve the performance of the rule-base, therefore a modification of the original PRD is proposed. Some PRD variants based on that modification are developed and evaluated through their applications to elevator operation problems.

DOI： 10.3233/978-1-61499-828-0-54

Scopus

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：IEICE-INST ELECTRONICS INFORMATION COMMUNICATIONS ENG  

In this paper, the authors propose an integer linear programming (ILP) model for static multi-car elevator operation problems. Here, "static" means that all information which make the behavior of the elevator system indeterministic is known before scheduling. The proposed model is based on the trip-based ILP model for static single-car elevator operation problems. A trip of an elevator is a one-directional movement of that elevator, which is labaled upward or downward. In the trip-based ILP model, an elevator trajectory is scheduled according to decision variables which determine allocations of trips to users of an elevator system. That model has such an advantage that the difficulty in solving ILP formulations resulted by that model does not depend on the length of the planning horizon nor the height of the considered building, thus is effective when elevator trajectories are simple. Moreover, that model has many variables relevant to elevators' positions. The proposed model is resulted by adding 3 constraints which are basically based on those variables and make it possible to prevent elevators in a same shaft from interfering. The first constraint simply imposes the first and last floors of an upper trip to be above those of its lower trip. The second constraint imagines the crossing point between upper and lower trips and imposes it ahead of or behind the lower trip according to their directions. The last constraint estimates future positions of elevators and imposes the upper trip to be above floors of passengers on the lower trip. The basic validity of the proposed model is displayed by solving 90 problem instances and examining elevator trajectories generated from them, then comparing objective function values of elevator trajectories on a multi-car elevator system with those on single-car elevator systems.

DOI： 10.1587/transfun.E100.A.385

Web of Science

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：Japanese   Publishing type：Research paper (scientific journal)   Publisher：Institute of Electrical Engineers of Japan  

In this paper, the author reports a study into usefulness of polysemy in the 6-multiplexer problem. The study is based on integer linear programming models for the problem of obtaining optimal classifiers for that multiplexer problem. Two classifier designs are considered each of which determines outputs according to inputs and classifiers. One design is typical. In this design, one classifier has one action, and one classifier supports only one output. In the other design with polysemy, one classifier has votes for all actions, and one classifier may support multiple outputs. In both designs, majority voting by matching classifiers determines the output for the corresponding input. Integer linear programming models are developed for some problem settings which differ in the classifier design, the number of classifiers, and the usage of the default rule which does not care the input. Solving those models display that if the number of classifiers is 4 and the default rule is not permitted, the design with polysemy yields more effective classifier sets than the other design.

DOI： 10.1541/ieejeiss.136.299

Scopus

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Authorship：Lead author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：WILEY-BLACKWELL  

In this paper, we propose an approach for classifying customers in retail stores into given types according to their shopping paths, each of which is a sequence of sections visited by the corresponding customer and is gathered by an RFID tag. The approach vectorizes a sequence of sections; that is, the approach splits such a sequence into tuples of sections, then sums up the occurrence counts of those tuples. This vectorization is based on the hypothesis that a customer's type has a relation to subsequences of sections in his/her shopping path and the conjecture that customers types can be attributed to co-occurrences of such subsequences. After vectorization, the proposed approach applies a general discrimination method to such vectors of equal length. In computational illustrations, principal component regression is selected as a representative of general discrimination methods and is applied to shopping paths collected in an existing retail store so as to predict whether a customer purchases items more than average or not. Computational results demonstrate the effectiveness of the proposed approach, with higher forecast accuracies than existing methods. (C) 2014 Wiley Periodicals, Inc.

DOI： 10.1002/ecj.11588

Web of Science

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：IJNC Editorial Committee  

In this paper, we propose a basic technique to minimize the computational time in executing the infinite-stage dynamic programming (DP) on a GPU. The infinite-stage DP involves computations to probe whether a value function gets sufficiently close to the optimal one. Such computations for probing convergence become obvious when an infinite-stage DP is executed on a GPU, since those computations are not necessary for finite-stage DPs, and hide behind loops for updating state values when a DP is executed on a CPU. The heart of the proposed technique is to suppress those computations for probing by thinning out them. By the proposed technique, differences between state values before and after being updated are periodically transferred to the main memory, then are checked to probe convergence. This intermittent probing makes contrast to ordinary methods in which computations for probing are processed every time. The technique also proposes a formulation to determine optimal periods for probing based on simple statistics given by preliminary experiments. The effectiveness of the proposed technique is examined on two problems; the one is a kind of the animat problem in which an agent moves around in a maze to collect foods, and the other is the mountain-car problem in which a powerless car on a slope struggles to pass over a higher peak. Computational results display that a method with the proposed technique decreases computational times for both problems compared to methods in which computations for probing convergence are processed every time, and the degree of decreasing seems remarkable when the state space is larger.

DOI： 10.15803/ijnc.4.2_321

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

In this paper, we propose a variant of the dynamic programming which is suitable for solving infinite Markov decision processes on GPUs. The primary feature of the proposed method is to not always but intermittently transfer and check values for proving the convergence of the procedure. It is expected for the proposed method to decrease computational times by suppressing surplus transfers and checks of values. This expectation is verified through applications of some dynamic programming programs to a simple animat problem and the mountain-car problem. © 2013 IEEE.

DOI： 10.1109/CANDAR.2013.44

Scopus

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

In this paper, we propose a technique for a trip-based integer programming model for elevator operations to decrease computational times in solving problems for multiple elevators. The technique is inspired by the speculation on ideal trajectories of elevators, where the interval of arriving floors is equal to each elevator. Such trajectories are regarded as oscillating waves. It is expected that limiting movements of elevators to resemble such waves may decrease computational times without degradation on the objective value. This expectation is numerically examined by obtaining optimal trajectories of some problem instances. © 2013 IEEE.

DOI： 10.1109/GCCE.2013.6664807

Scopus

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：Institute of Electrical Engineers of Japan  

In this paper, we propose an approach for classifying customers in retail stores into given types according to their shopping paths, each of which is a sequence of sections visited by the corresponding customer and is gathered by an RFID tag. The approach vectorizes a sequence of sections
that is, the approach splits such a sequence into tuples of sections, then sums up occurring counts of those tuples. This vectorization is based on the hypothesis that a customer's type has relation to sub-sequences of sections in his/her shopping path and a conjecture that customers' types can be attributed to co-occurrences of such sub-sequences. After vectorization, the proposed approach applies a general discrimination method to such vectors of equal length. In computational illustrations, the principal component regression is selected as a representative of general discrimination methods and is applied to shopping paths collected in an existing retail store so as to predict whether a customer purchases items much than average or not. Computational results display the effectiveness of the proposed approach as higher forecast accuracies than known works. © 2012 The Institute of Electrical Engineers of Japan.

DOI： 10.1541/ieejeiss.132.2051

Scopus

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

In this paper, we propose an integer programming formalization to optimize elevator operations under such an ideal condition that all information on passengers who arrive during the planning period is known in advance. The basic idea which inspired the formalization is the concept of trip, which represents a uni-directional (upward or downward) movement of an elevator and forms a trajectory of an elevator with other trips. In the formalization, all elevators are alloted to equal number of trips, and a passenger is assigned to one of trips of the same direction to that passenger. The number of necessary trips is unknown without an optimal schedule, whereas an optimal schedule can not be obtained without that number. This complication is straightforwardly solved by the incrementally resolving procedure by which the number of available trips starts with two and is incremented by two until an increment does not yield a more effective schedule. A supposition on the working of that procedure leads to the gradually resolving procedure, which restricts possible trips of passengers within neighborhoods of trips which are optimal under a fewer number of available trips. In computer illustrations, some problems which differ in numbers of elevators, passengers, and traffic patterns are considered. A certain number of embodied problem instances are generated for each problem, and an optimal schedule is obtained for each problem instance. Computational results display that the number of available trips seems to primarily affect computational costs, and a problem with multiple elevators can be easier as expected since such a problem does not require so many trips in usual. Additionally, the effectiveness of the gradually resolving procedure is shown as it yields schedules optimal in most cases within radically shorter computational times. © 2012 IEEE.

DOI： 10.1109/SCIS-ISIS.2012.6505108

Scopus

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

This paper presents the details of a fundamental study of analyzing a customer trajectory by clustering the constituents of the customer trajectory sequence into two types: one, the constituents working as a source of the features of a group of customer trajectory sequences in a given period that we call a time slot and the other, the constituents working as a source of personal features of each of the customers. Removing personal features from a group of customer trajectory sequences provides us new knowledge that is less affected by the difference among sampled data and is expected to be utilized for planning sales strategies that target customers shopping in specific time slots and for evaluating the effectiveness of these strategies. In this paper, we introduce definitions and procedures of the proposed method of clustering the constituents of customer trajectory sequences. © 2010 ICIC International.

Scopus

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：JAPAN SOC MECHANICAL ENGINEERS  

In this paper, an approach to complement legacy rules for the elevator operation is proposed. The approach is derived from the analysis that the elevator operation in the real world often obeys a heuristic rule, and such a rule can be divided into a legacy rule and ad-hoc rules. In the approach, ad-hoc rules are represented as polysemous rules, and a Genetics-Based Machine Learning (GBML) method is applied to acquire such rules. Here, a polysemous rule encodes, not a set of environments' states as the well-known if-then rule does, but a relative attribute vector of an arbitrary elevator. The elevator selection rule based on polysemous rules is simple: if there is a polysemous rule which matches one of attribute vectors of the elevators, select the elevator which corresponds to the matching vector; otherwise select an elevator according to a legacy rule. In computer illustrations, the GBML method is applied to 3 traffic patterns formed by the system's users. It is shown that the resultant polysemous rules seem to complement an existing (legacy) operational rule. Furthermore, polysemous rules, which are selected among those acquired by the GBML method, are successfully applied to harder problems with more elevators than those used in learning.

DOI： 10.1299/jamdsm.4.651

Web of Science

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：The Society of Instrument and Control Engineers  

In this paper, we present a modified dynamic programming (DP) method. The method is basically the same as the value iteration method (VI), a representative DP method, except the preprocess of a system's state transition model for reducing its complexity, and is called the dynamic programming on reduced models (DPRM). That reduction is achieved by imaginarily considering causes of the probabilistic behavior of a system, and then cutting off some causes with low occurring probabilities. In computational illustrations, VI, DPRM, and the real-time Q-learning method (RTQ) are applied to elevator operation problems, which can be modeled by using Markov decision processes. The results show that DPRM can compute quasi-optimal value functions which bring more effective allocations of elevators than value functions by RTQ in less computational times than VI. This characteristic is notable when the traffic pattern is complicated.

DOI： 10.9746/jcmsi.2.213

CiNii Books

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

In this paper, an approach using a Genetic-Based Machine Learning (GBML) method to acquire rule sets which complement legacy rules for the elevator operation is shown. A rule set is designed to be comprised of unit-rules, each of which encodes an elevator's augmented attribute vector and matches an arbitrary elevator which has such attributes. The GBML method is built in the Pittsburgh-style and tailored for acquiring such rules sets. In computer illustrations, the effectiveness of acquired rule sets is displayed as for three traffic patterns.

CiNii Books

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：Institute of Electrical Engineers of Japan  

This paper shows that it is possible to decrease the computational cost of Dynamic Programming methods for problems of planning a certain kind of Stochastic Discrete Event System (SDES), if those methods are constructed based on the modeling method called Symbolic Binary Coding Scheme (SBCS). The considered property of such a system is that the number of following states from an origin state is rather smaller than the number of possible situational-inputs (SIs) which the system suffers. Here, the SI is one of the components involved in a model by the SBCS, and a kind of augmented disturbance. The property raises the conjecture that some SIs may have the same influence to the system, and if so, they can be united to an SI which typifies them. This conjecture can be achieved by algebraically simplifying the state transition function, which is modeled as a set of elementary binary functions by the SBCS, with the aid of Computer Algebra Systems. That unification obviously decreases the computational cost of DP methods. In computer illustrations, the effectiveness of employing that unification is shown through results by applying the Value Iteration method to elevator operation problems of small scale. © 2009 The Institute of Electrical Engineers of Japan.

DOI： 10.1541/ieejeiss.129.1237

Scopus

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：Japanese   Publishing type：Research paper (scientific journal)   Publisher：The Society of Instrument and Control Engineers  

In the present paper, a dynamic programming (DP) approach for elevator operation problems is reported, in which two kinds of the design of DP algorithms are introduced. One (Value Iteration; VI) is rather naive implementation of DP concept to the problems, and the other (Dynamic Programming on Reduced Models; DPRM) is an improved version in computational efficiency by reducing the accuracy of the utilized state transition model. The basic idea of this reduction is as follows: it is expected that dropping some situational-inputs, which represent the possible causes for each state transition, with low occurring probabilities may reduce the size of the state transition model without losing the essence of a system's behavior. In computational illustrations, the VI and the DPRM are applied to two types of the problems of small sized, and it is observed that the computational time required in applying the DPRM is much less than that in applying the VI for each problem, while the results obtained by both methods are almost equal with respect to not only averaged but also maximum waiting time over passengers. Furthermore, the optimal and quasi-optimal policies obtained by the VI and DPRM are compared with an existing elevator allocation rule (Call Dispatching; CD), and the results show both policies are statistically better than the CD.

DOI： 10.9746/ve.sicetr1965.44.174

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：Japanese   Publishing type：Doctoral thesis  

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

In this paper, a discrete model of an elevator system and a design of dynamic programming (DP) method based on that model are shown. The method is followed by a modified DP method on approximated state transition models. The discrete model theoretically considers the causes of a system's probabilistic behavior, and leads such approximation by the reduction of the insignificant constituent of them. In computational illustrations, the two DP methods are applied to two problems of small scale. The results show that the policies obtained by DP methods are both statistically better than an existing elevator allocation rule, and the modified method requires less computational time than the another. © 2007 SICE.

DOI： 10.1109/SICE.2007.4421439

Scopus

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：Japanese   Publishing type：Research paper (scientific journal)   Publisher：The Institute of Electrical Engineers of Japan  

In this paper, we show an approach to acquire a rule which is capable of achieving effective elevator services. The allocation of elevators to hall calls is taken as the decision variable and the Pitt approach of a Genetics-Based Machine Learning (GBML) method is designed according to the representation and the characteristics of the problem. We use a rule architecture in which a combination of some measures is directly encoded as a unit-rule and an elevator allocation procedure is represented as a rule set composed of some unit-rules. Five measures are developed based on the results of the preliminary experiment and it is shown that the rule as the weighted summation of the measures (called WS) is more effective than a conventional elevator allocation rule. In computational experiments, the up-peak, down-peak and two-way traffic patterns are considered and three rule sets are acquired by applying the GBML method. Based on the results, we can confirm that the GBML method can acquire some unit-rules appropriate for the targeting traffic patterns, and we can mention that, for some presumed traffic patterns, the possibility of constructing an effective rule set by combining some specific rule sets.

DOI： 10.1541/ieejeiss.126.761

CiNii Books

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：Japanese   Publishing type：Research paper (bulletin of university, research institution)  

CiNii Books

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：Japanese   Publishing type：Research paper (scientific journal)  

In this paper, we propose a framework for obtaining the optimal car service to elevator operation problems by applying branch-and-bound methods based on the deterministic optimization model. In building the model, we assume the followings: the numbers such as time and car positions axe discritized, the movement of cars and passengers synchronized with discrete time, and all passengers arriving to the hall is known beforehand. In the model, the transportation of any passenger is considered as a combination of two jobs, i.e., an into-job and an out-of-job. The into-job corresponds to a passenger's getting into a cax, while the ont-of-job corresponds to getting out of a car. Here, the optimal car service of the problem is obtained by assigning each into-job to an appropriate car and determining the processing order of into- and out-of-jobs for each car under some constraints including the precedence conditions, In designing a BAB solution, the assignment of into-jobs to cars and the processing order of jo bs on each car are taken as decision variables. It is expected that the optimal (or near-optimal) car service obtained by applying such techniques as branch-and-bound methods based on the model is helpful to estimate the effectiveness of the utilization of look-ahead information. It is also useful in valuating the performance of the existing rules for elevator operation. In the paper, by using the proposed method, the validity of an existing rule, i.e., the Call-Dispatching and Selective-Collective rule (CDSC), is examined. The results of computational experiments show that the performance of the CDSC rule is not always the optimal or near-optimal, while jt reveals a practical potential, i.e., it generates rather good car services within very short time. © 2003, The Institute of Electrical Engineers of Japan. All rights reserved.

DOI： 10.1541/ieejeiss.123.1334

Scopus

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Authorship：Lead author,　Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Event date： 2021.9

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Event date： 2021.9

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Event date： 2021.5

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Event date： 2021.5

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Event date： 2020.11

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Event date： 2020.11

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Event date： 2020.9

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Event date： 2017.5

Language：Japanese  

researchmap

Event date： 2014.7

Language：Japanese  

researchmap

Event date： 2013.7

Language：Japanese  

researchmap

Event date： 2006

researchmap

Event date： 2005

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Event date： 2001.9

Language：Japanese  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Akita, Japan  

In this paper, we show an approach to acquire a rule set which has a capability of achieving effective elevator movements. It is expected that an efficient schedule is obtained by properly determining the floors of each elevator is going to move in the next time whenever an event occurs in an elevator system. We select such direct assignment of floors as the decision variables to schedule elevator movements in expectation of realizing the fore-waiting elevator movements. Each of the rules represents the values of parameters used in the function which is defined to estimate a cost of selecting a determination, i.e., the floors to which elevators should move if no additional event occurs after the previous one. A method to acquire effective rule sets is constructed based on the Pitt approach of Genetics-Based Machine Learning. Preliminary results by using the method and further points to be considered are shown.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

In this paper, we propose a computerized method to support solving problems with relative evaluation, for which we can not evaluate individual solution candidates but can evaluate by comparing one with others. Designing a pattern generator for a walking robot can be thought as one of such problems since one can intuitively know which is a better walking pattern than another but it is impossible to give the consistent evaluation of the goodness for individual walking patterns. One feasible approach inspired by one research about the human creativity done in the field of cognitive science is to seek the evaluation itself during the search for the solution for walking robot. A coevolutionary GA is used to implement the approach and is applied to the problem designing a pattern generator for a biped locomotive robot.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

In this paper, we present a computer-aided approach to designing a pattern generator for a biped robot. The approach is based on a semiotic model of human creativity. By the model, the design creative process is seen as a composition of morphisms between design spaces that can computationally be interpreted in a co-evolutionary framework as two searches: one in a fixed problem space for problem solutions, and another --- among different problem spaces for the most ``adequate'' space. A software system is built and a case study is made.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Kyoto, Japan  

In this paper, we propose a framework for obtaining the optimal car service to elevator operation problems by applying a branch-and-bound method based on the deterministic optimization model. In designing the model, it is assumed that the movement of cars and passengers synchronized with discrete time, and all passengers arriving to the hall is known beforehand. The assignment of passengers to cars and the processing order of the services to passengers are taken as decision variables in designing a branch-and-bound solution. In the computational experiments, the Call-Dispatching and Selective-Collective rule is taken as a typical rule and the results obtained by using this rule and those by the branch-and-bound method are compared and examined.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：English   Presentation type：Poster presentation  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Poster presentation  

researchmap

Language：Japanese   Presentation type：Poster presentation  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Hyogo (canceled)  

In this paper, we display an approach for acquiring rule sets in consideration of the cooperation from experts. The approach is based on a Genetics-Based Machine Learning method and a simple design of the rule set. The design is unique in the sense that a constituent of a rule set does not form an if-then rule as usual, but a such-that rule which indicates a relative attribute vector of a taken decision. The usability of the approach is roughly examined through a preliminary application to a well-known static job-shop scheduling problem.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Osaka, Japan  

In this paper, results by applying a genetics-based machine learning method to acquire approval voting polysemous rule sets (AVPRSs) for elevator operation problems are shown. The design of the AVPRS has been devised for a job-shop scheduling problem to decrease the arbitrariness in selecting decisions when multiple candidates are indicated by matched polysemous rules. An AVPRS elects a decision approved by most polysemous rules in such a case, and is expected to function well even when individuals encode fewer polysemous rules. This expectation is examined through computer illustrations for some settings which differ in the probability that an element of a polysemous rule to mutate the DON'T CARE symbol.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Shiga, Japan  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Kyoto, Japan  

In this paper, we propose a technique to improving the efficiency of a genetics-based machine learning method in which individuals are evaluated on many problem instances. The technique is based on records of problem instances hardest for individuals' direct ancestors. An individual is firstly evaluated on such a problem instance. If the resultant objective value is worse than that of the individual's ancestor, then the evaluation procedure is terminated and that individual is given its ancestor's fitness. In computer illustrations, the effectiveness of the technique is examined through its application to elevator operation problems with some pseudo random numbers.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Osaka, Japan  

In this paper, state transition probabilities of an elevator system (ES) are formalized based on a discrete model of that ES. This model considers the causes of a system's probabilistic behavior, which are called situational-input. The probability of a state transition is represented as the summation of rising probabilities of situational-input which bring such transition. A dynamic programming (DP) method is designed on the basis of those constituents. This is a configuration of value iteration algorithm, where the immediate cost is the number of registered hall calls at one moment. It is followed by a modified DP method (DPRM) that is based on reduced state transition models. Replacing the whole situational-input space by smaller and essential space causes such reduction. If most of situational-input have low rising probabilities, DPRM can result cost functions which are not much less effective than those by ordinary DP methods in shorter computational time. In computational illustrations, two DP methods are applied to a problem of small scale. The result shows that optimal policies are significantly more effective than an existing elevator allocation rule, and they are not so different in terms of passengers' averaged waiting times.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Hyogo, Japan  

In this paper, we show an application of a straightforward elevator allocation rule based on approximate values of state transition probabilities. The values are naively calculated by making use of the state transition function which clearly manifests the Markov property of the decision making process on a descritized elevator system model. A state of the system is represented as a combination of boolean variables and the state transition function symbolizes the simultaneous equations, where the number of them is equal to the number of the boolean variables. Some preliminary comparisons of elevator movements by a rule which assigns an elevator with a minimum expected pseudo wait time to those by a legacy rule are shown on a problem of small scale.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Tokyo, Japan  

In this paper, we make a discrete model of an elevator system and formalize state transition probabilities based on the model on the assumption that the traffic pattern is stationary in the planning time. An appropriate function which determines the behavior of the model can be inferred by the fact that the elevator operation has Markov property when unmeasurable and/or uncontrollable influence by passengers is explicitly considered. State transition probabilities are formalized based on the function and its availability is shown on a problem of small scale through the evaluation of a straightforward car allocation rule using the probabilities.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Ibaraki, Japan  

In this paper, we show a model of the elevator system which consists of one elevator. The state space of the elevator system is represented as a combination of many binary variables in the model. Two existing Q-learning methods, (i) one is a simple method with a greedy agent, and (ii) the other is such that simultaneously updates a set of Q-tables and can adaptively focus on the Q-table on which the states are effectively discriminated. Two reward functions are also considered and it is noted that using the method (ii) with the reward as the change of the number of registered hall calls may provide an insight to overcome the SC rule, a representative to calculate the command for an elevator.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Hokkaido, Japan  

In this paper, a simple rule-base design for elevator operations is displayed. On the design, a unit-rule which comprises a rule-base corresponds to an elevator's characteristics vector which is normalized among all elevators. By using such a rule-base, if a characteristics vector of an elevator matches a unit-rule, such elevator is selected; if not, a default rule is used to select an appropriate elevator. This simple design may: (i) make it easy to learn effective chromosomes by decreasing the epistatis between genes, and (ii) make it possible to combine some rule-bases specific to some environments into a comprehensive rule-base. These expectations are examined through preliminary computer illustrations.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Hyogo, Japan  

This paper shows a simple parallelization of computing reachable sets of stochastic discrete event systems (SDESs) which exhibit the imbalance in the number of states following to origin states. The estimation of the computational cost of following states becomes possible by: (i) explicitly formalizing state transition functions of such SDESs by means of the Symbolical Binary Coding Scheme, then (ii) algebraically simplifying those functions. It is expected that distributing origin states according to that estimation in a greedy fashion is more effective than doing so in a round-robin fashion. This expectation is examined through computational illustrations as to an elevator system of small scale, which is an example of such a SDES.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Kyoto, Japan  

This paper shows computational results of parallelizing the computation for assembling reachable sets of elevator systems. By employing the state transition function, which is formalized by the binary coding scheme, of an elevator system, the calculation of states which follow to a seed state is represented as a simple and independent formula. That calculation is plainly parallelized by distributing seed states to processors in a round-robin fashion. In computational illustrations, computational times for assembling reachable sets of an elevator system of small scale are shown.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Hokkaido, Japan  

This paper shows that algebraically simplifying state transition functions leads to the decrease in computational cost in applying Dynamic Programmings (DPs) to problems of scheduling Stochastic Discrete Event Systems (SDESs). When using a model by the employed binary coded modeling scheme for SDESs, computational cost of DPs is roughly proportional to the size of a situational-input space, which is a kind of augmented disturbance space. In such model, the state transition function is formalized as a set of many binary functions, and they can be represented as expressions on Computer Algebra Systems (CASs). The complexity of those expressions can be decreased by algebraically localizing them, i.e., limiting their domains to a specific sub state space (this is equivalent to fixing the corresponding state variables in the expressions), then applying simplification rules to them on a CAS. On the resulting expressions, some situational-input variables do not affect the state transition, and removing those variables makes smaller situational-input spaces, each of which is localized to the corresponding sub state space. The effect of using these smaller spaces instead of the original is examined through computational results by applying the Value Iteration method to elevator operation problems.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Osaka, Japan  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

In this paper, two approaches which have been taken by the author to the elevator operation problem are introduced. The aim of the one approach is to obtain optimal solutions regardless of the scale of applicable problems, and that of the other is to obtain effective solutions regardless of their optimality. The former approach yields a modeling scheme for stochastic discrete event systems, then brings optimal elevator allocation rules by the dynamic programming. The latter approach yields a rule design applicable to large systems, then makes it possible to acquire effective elevator allocation rules by the genetics-based machine learning.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Kyoto, Japan  

In this paper, we propose to adopt the ``majority rule'' into a Genetics-Based Machine Learning (GBML) method, which had been proposed by the authors to acquiring polysemous rules, so as to cope with situations where multiple rules are matched. By the majority rule, counts of which polysemous rules match to decisions are calculated, and such a decision of a maximum on matching counts is selected. In computer illustrations, results by applying a legacy GBML method and a GBML method with the majority rule to a static job-shop scheduling problem are displayed on some parameter settings which mainly differ in the number of polysemous rules.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Kyoto  

In this paper, we display an approach to effectively acquire polysemous rule sets for job-shop scheduling problems. The approach is straightforward and as follows: (i) dividing a target problem into sub-problems, (ii) applying a machine learning method for each sub-problem, (iii) gathering resultant rule sets into a seed set, (iv) again applying the method to the original problem with rule sets initialized from the seed set. Some computational results by the approach are roughly examined through its application to a well-known static job-shop scheduling problem.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Osaka, Japan  

In this paper, we preliminarily examine the effectiveness of Action-Less Rule Sets (ALRSs) as rule-bases for representing experts' knowledge. The ALRS is expected to be suitable for such purpose, since it can cover various situations by small number of unit-rules, and those unit-rules can be easily interpreted by experts. In computer illustrations, the knowledge that what characteristics should be considered in selecting next operations in a job-shop scheduling problem is manually added by the decision maker according to Gantt charts, and values of characteristics are learned by a genetics-based machine learning method.

researchmap

Language：English   Presentation type：Oral presentation (general)  

Venue：Kyoto, Japan  

In this paper, we display an illustration of classifying customers in a retail store by using stream data gathered by RFID tags and POS data. In the illustration, a customer is grouped according to a set of binary variables each of which shows whether the section which corresponds to that variable was visited before, or after, main-dish sections. Computational results display that a simple conjecture about customers' shopping paths is supported by the resultant classification, and it is possible to guess the behavior of customers of a specific type and propose a plan for improving the layout of sections.

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

Venue：Aomori, Japan  

The mountain-car problem and an animat problem on a slippery plane are selected as representatives of deterministic and stochastic MDPs, respectively.
In implementing DPs, the CUDA is selected as a framework to write and execute general-purpose programs on GPUs.
The efficiency of programs on the CUDA is strongly affected by the memory access, which is desired to be few in number and synchronized between cores in timing.
Thus, such a dialect of programs is also considered that terminates according not to the maximum progress of value functions but the iteration count directly specified.
Computational illustrations show that basic speedups are possible.
It is also implied that the decrease on the memory access brought by directly specifying the iteration count and the regularity in accessing memory found in the animat problem yield further speedups.

researchmap