In this paper, a cooking menu is created by combination of some dishes. The combinatorial optimization problem is considered well-balanced for nutrition, matching of dishes and the number of dishes in each group. In order to create a well-balanced menu, we need to take account of the intakes of nutrients, energy, fat, dietary fiber, protein, carbohydrate, salt and etc. They have ideal value or range toward each person. It is difficult for combination of some dishes to satisfy the value or range. Usually, required intakes of each nutrient are not rigid strongly. Therefore, they are expressed by fuzzy numbers and membership functions are set to appropriate form. It aims at maximization of the minimum membership function value to take the balance for nutrient. The factors of matching and group composition are expressed in restriction conditions. This problem is formulated by fuzzy mathematical programming. Effectiveness of the method is presented by numerical experiments with 180 dishes data.

View full abstract

In this paper, we propose a location based internet advertisement model and its application to an advertisement system for mobile phone using GPS modules. In general, one of the purposes of the Internet advertisement systems is to set the advertising target to users whom the advertisement is effective to. Nowadays location information of users can be easily obtained through various internet services. Therefore, in the proposed model, location information is used for user targeting. A basic idea of the model is to show an advertisement to users near the target area of the advertisement. By using our improved neural network, effective area of the advertisement is automatically predicted through the trials of advertisement presentations. We develop a new advertisement system using this model. The experiments of the system show the effectiveness of the proposed model.

View full abstract

A risk estimation system based on Bayesian Networks using a driver model is proposed to achieve a collision evasion function and a risk estimation function in traffic situations for safe driving support systems. The proposed system uses input traffic and vehicle information to evaluate subsequent driving operations, such as acceleration and steering, using Bayesian networks. The vehicle trajectory is then forecasted using a dynamical physical model. Next, the risk of collision with other vehicles is calculated based on the output probabilities of the Bayesian network and the predictions of the dynamical model by considering the trajectories of other vehicles and the possible trajectories of the car itself. In the scene of the intersection, the effectiveness of the proposed system is shown by comparing the simulations for the scene where the accident occurs and for the scene where it does not, and then specifying the difference between the risk transition coefficients for both cases. Also, it is shown that the risk decreases after performing an evasive action. In addition, the proposed system is applied to a real environment data. In the future, the proposed system can be applied to a system that prevents traffic accidents by giving the optimal evasion driving operation to the automatic control system of the vehicle.

View full abstract

The methodology of state space construction is an important point in the application of reinforcement learning to real tasks. Self-organizing map (SOM) is a useful tool for state space construction. Furthermore, profit sharing (PS) is a practical learning method in a typical class of Partially Observable Markov Decision Processes (POMDPs). However, state space construction by SOM has been used only for Q-learning. In this paper, we present arrangement in a neighborhood function for adaptation of SOM to PS. The effectiveness of the proposed method was verified using a simulation experiment in a maze problem.

View full abstract

A car navigation system predicts the destination by using the driving history, including such factors as driving time and route. Depending on the operating factors used for limiting the driving history, the effectiveness of calculating the destination differs. This paper evaluates the difference in the driving factors used in determining the destination. The effectiveness of destination prediction was evaluated by using the mutual information between the driving factors and destination. The mutual information increases when the number of histories is reduced. The mutual information increases too much when the driving factors have very little history only 1 or 2 examples. We evaluated using the mutual information normalized by eliminating the change information by the number of histories. Driving factors were evaluated using 20 users driving histories over approximately a 3-month period. As a result, the driving time factor was found to be effective in calculating destination prediction immediately after the user starts to drive. The driving route factors, such as starting place and 2 main crossings through which the user passes first and second, were effective in calculating the destination during the user's drive. Therefore, the technique of destination prediction using the driving history, which time is the same as the predicted drive immediately after the user started to drive and the driving history, which route is the same as the predicted drive during the user's drive, is effective.

View full abstract

In this paper, two clustering algorithms that handle data with tolerance are proposed. One is based on hard c-means (HCM) while the other is based on the learning vector quantization (LVQC). We consider a tolerance which is a new concept to handle data with uncertainty such as errors, ranges, or a lost attribute of data in the optimization framework. The concept of tolerance is included in both algorithms. Dissimilarity in the former clustering algorithms is defined by using nearest-neighbor, furthest-neighbor or Hausdorff distance. On the other hand, dissimilarity in the proposed algorithms is defined by squared L₂ (euclidean)-norm and the algorithm can handle the data with uncertainty in the strict optimization problems. First, the concept of tolerance which implies errors, ranges and the loss of attribute of data is described. Optimization problems that take the tolerance into account are formulated. A unique and explicit optimal solution is given by Karush-Kuhn-Tucker conditions. An alternate minimization algorithm and a learning algorithm are constructed. Moreover, effectiveness of the proposed algorithms is verified through numerical examples.

View full abstract

In this paper, focusing on nonlinear programming problems involving constraints, we attempt to propose a general-purpose and high-performance approximate solution method for them. In recent years, as general-purpose approximate solution methods for nonlinear programming problems, the particle swarm optimization (PSO) method has drawn considerable attention. However, there exist few reports with respect to successful results for constrained nonlinear programming problems by PSO-based methods. Furthermore, the PSO-based methods have a shortcoming that the search is liable to stopping at a certain local solution. Thereby, we incorporate the bisection method and a homomorphous mapping to carry out the search considering constraints. In addition, we incorporate a multiple stretching and secession into PSO method for restraining the stopping at a certain local solution. Furthermore, we show the efficiency of the proposed revised PSO method (rPSO) by comparing it with two existing methods such as GENOCOPV and αPSO through the application of them into the numerical examples.

View full abstract

The centrality of network is an index to know whether what kind of node and link are important in the network. In this paper, we produce an expansion model of network generation based on the previous model, where all agent in the network select a new link by voting based on own centrality in consideration for the network in which its link is added.There are two contributions in this paper: we expand the model to use multiple centrality, another contribution is that we estimate the rational of agents network. The former is that we assume that some kinds of centralities affect the topology of network, so that we consider each centrality to be an attribute, and use an addition of utility rule in case of vote. The later is that our model is seen as a new methodology to analyze the causes to consist a complex network. As a result, the combination of centrality is effective for generation of various networks, and showed new approach of social network analysis by application of our proposed model.

View full abstract

In this paper, we propose a chaotic complex-valued associative memory which can realize a dynamic association of multi-valued patterns. The proposed model is based on a complex-valued associative memory and a chaotic associative memory. The complex-valued associative memory can treat multi-valued patterns, and the chaotic associative memory can recall stored patterns dynamically. The proposed model utilizes the properties of these conventional models to realize the dynamic association of multi-valued patterns. In this paper, we showed that the chaotic behavior occurs in the proposed chaotic complex-valued neuron model. Moreover, we carried out a series of computer experiments and confirmed that the proposed model realizes the dynamic asscotiation of multi-valued patterns. And, we discussed the influence of some parameters and state number for dynamic associations.

View full abstract