Catálogo de publicaciones - libros

In this paper, we demonstrate that simple fuzzy rule-based classification systems with high interpretability are obtained through multiobjective genetic rule selection. In our approach, first a prespecified number of candidate fuzzy rules are extracted from numerical data in a heuristic manner using rule evaluation criteria. Then multiobjective genetic rule selection is applied to the extracted candidate fuzzy rules to find a number of non-dominated rule sets with respect to the classification accuracy and the complexity. The obtained non-dominated rule sets form an accuracy-complexity tradeoff surface. The performance of each non-dominated rule set is evaluated in terms of its classification accuracy and its complexity. Computational experiments show that our approach finds simple fuzzy rules with high interpretability for some benchmark data sets in the UC Irvine machine learning repository.

When the data is given as mixed data, that is, the attributes take the values in mixture of binary and continuous, a clustering method based on -means algorithm has been discussed. The binary part is transformed into the directional data (spherical representation) by a weight transformation which is induced from the consideration of the similarity between binary objects and of the natural definition of descriptive measures. At the same time, the spherical representation of the continuous part is given by the use of multidimensional scaling on the sphere. Combining the binary part and continuous part, like the latitude and longitude, we obtained a spherical representation of mixed data. Using the descriptive measures on a sphere, we obtain the clustering algorithm for mixed data based on k-means method. Finally, the performance of this clustering is evaluated by actual data.

This paper presents several fuzzy clustering methods based on self-organized similarity (or dissimilarity). Self-organized similarity (or dissimilarity) has been proposed in order to consider not only the similarity (or dissimilarity) between a pair of objects but also the similarity (or dissimilarity) between the classification structures of objects. Depending on how the similarity (or dissimilarity) of the classification structures cope with the fuzzy clustering methods, the results will be different from each other. This paper discusses this difference and shows several numerical examples.

The amount of data is growing at an exponential rate. We are faced with a challenge to analyze, process and extract useful information from the vast amount of data. Traditional data analysis techniques have contributed immensely in the area of data analysis but we believe that the soft data analysis techniques, based on soft computing techniques, can be complementary and can process complicated data sets. This paper provides an introduction to the soft data analysis paradigms. It summarizes the successful and possible applications of the soft computing analysis paradigms. The merits and demerits of these paradigms are included. A number of resources available are listed and the future vision is discussed. This paper also provides a brief summary of the papers included in the session on “Innovation in Soft Data Analysis”.

Physiological experiments demonstrate establishment of immunological tolerance is controlled by dose-timing and dose-amount of an antigen. My previous study reproduced the tolerance dependent on dose-timing of an antigen in a two Bcell clones network model with an “asymmetric” Bcell-Bcell interaction. This study first clarifies its mechanism using a dynamical system technique:nullcline analysis. Next, this study proposes a three Bcell clones network model with the same style of Bcell-Bcell interaction, and shows the model simulation can reproduce the tolerance dependent on dose-amount of an antigen: high and low zone tolerance. The success of this reproduction is worthy of attention. Because theoretical studies based on the traditional “symmetric” immune network modeling scheme have not been able to reproduce it well. This study would teach us the renewed recognition of “asymmetric” immune network modeling scheme.

We have applied our previous immunity-based system to anomaly detection for network traffic, and confirmed that our system outperformed the single-profile method. For internal masquerader detection, the missed alarm rate was 11.21% with no false alarms. For worm detection, four random-scanning worms and the simulated were detected with no missed alarms and no false alarms, while a simulated was detected with a missed alarm rate of 80.57%.

In our previous studies, the immunity-based diagnostic model has been used by stationary agents in linked networks or by mobile agents on wired computer networks. We have not yet analyzed the performance of the diagnosis in wireless network where agents can move freely. In this paper, the diagnosis is applied to static and mobile sensor nodes in a 2-dimensional lattice space for wireless sensor network. Some simulation results show the strategy of going straight in the different direction can have the best detection rate. In addition, when the fraction of mobile nodes is changed, the transitions of the detection rate for the migration strategies are different.

This paper reports a multiagent approach to a basic model inspired by the asymmetric war between HIV and T-cells. The basic model focuses on the asymmetric interaction between two types of agents: Virus Agents (abstracted from HIV) and Immune Agents (abstracted from T-cells). Virus Agents and Immune Agents, characterized respectively as “generalists” and “specialists”, may be compared with asymmetric wars between computer viruses and antivirus programs, between guerrillas and armed forces, and so on. It has been proposed that antigenic diversity determines the war between HIV and T-cells. We also formalize the diversity of “generalists” that would determine whether generalists or specialists won. The multiagent simulations also suggest that there is a diversity threshold over which the specialist cannot control the generalist. In multiagent approaches, two spaces, Agent Space and Shape Space, are used to observe not only the spatial distribution of agent populations but also the distribution of antigenic profiles expressed by a bit string.

This paper reports on the construction and use of a dynamic relational network that has been studied as an immunity-based system based on the concept of immune networks. The network is constructed by an immunological algorithm that tunes the network not to react to normal sensor data, but to react to abnormal data thereafter. The tuning is not straightforward, since even normal sensor data involve many situations such as accelerating phase and cruise phase; on-road and off-road; and running at altitude. A case study on sensor systems for the combustion control system of an automobile engine is presented.

We introduce an interaction scenario in which users of public places can see relevant information items on public displays as they move. Public displays can dynamically collaborate and group with each other so as to minimize information clutter and redundancy. We analyse the usability constraints of this scenario in terms of information layout on the screens. This allows us to introduce a decentralized architecture in which information screens as well as users are modeled by software agents. We then present a simulator that implements this system.