122 J.R. Shulcloper, "Logical Combinatorial Pattern Recognition," Foro Iberoamericano de Reconocimiento de Patrones, pp. 123-138, Barcelona, Spain, September 2000. Logical Combinatorial Pattern Recognition José Ruiz-Shulcloper IRIS Lab. Electrical and Computer Engineering Department, The University of Tennessee, Knoxville, USA. [email protected] Laboratory of Pattern Recognition, Institute of Cybernetics, Mathematics and Physics, Havana, Cuba. [email protected] Abstract. In this paper we summarize the main results obtained by the Logical Combinatorial Pattern Recognition (LCPR) Group. The characteristic features of our approach and the relationship with the other approaches to Pattern Recognition (PR) are explained in a framework of a unique science. This approach was introduced originally in the Soviet Union under the orientation of the Academic Yuri Ivanovich Zhuravlev from the Computer Center of the Russian Academy of Sciences. 1. INTRODUCTION In our point of view, PR is a science with a strong interdisciplinary and applied character. Dealing with engineering, mathematical and computer processes over physics (photos; holograms; handwritings; hieroglyphics; symbols; bioelectrical signals; acoustic signals; etc.) and/or abstract objects (n-tuples of a certain Cartesian product of sets of several kinds: hard, fuzzy, rough,...) with the purpose (by computer devices and/or humans) to obtain the information that allows the establishment of properties and/or relationships of certain subsets or between subsets of the universe of objects. Based on this, we show in Table 1 a schema of scope of the PR problem. It is not in any case exhaustive. With this schema we want to underline that PR is an interdisciplinary science, that it has strong connections with mathematics, engineering and computer sciences. Also we want to illustrate the relationship between its component parts and its relative position with respect to those disciplines. Almost all the researches in this field are applied researches or at least with a big concern with the applications of its results. That is, PR is essentially an applied science. Nevertheless, that does not mean that there are not fundamental researches in this field. Perhaps is the time to remember, a statement by K. Levin “Rien n’est aussi practique qu’une bonne theorie”. Historically PR had being fractioned in small parts, showed in the schema, which we believe that is important do not consider as non-communicated areas. We think that we have an extensive area of researches and this area will increase while more interconnections we can to establish between all of those component parts. As we know there are two main forms of representation of objects in the framework of PR: in terms of certain alphabet of primitive parts of the objects, typical of the syntactic or structural approach; or in terms of certain set of features (variables), in an indirect way, typical of statistical and the rest of the approaches to PR. In order to explain what LCPR means, from here on we assume the latter approach to the representation for objects. The representation of objects usually is considered as a sequence of numerical or exclusively categorical values. Nevertheless, when we actually analyze the real world PR problems (feature selection, supervised or unsupervised classification problems), we found that in many cases it is not true that the description of objects taking into account for the specialists of these areas of knowledge is a flat description in terms of exclusively numerical or categorical features. TABLE 1. - Component schema of PR Mathematics Engineering Computer Mathematics Image Processing Signal Processing Image Analysis and Understanding Signal Analysis and Understanding Computer Vision Remote Sense Neural Networks for PR Genetic Algorithms for PR Artificial Intelligence Techniques for PR Mathematical Morphology Statistical PR Syntactical PR Logical Combinatorial PR There are many problems in which the description of the objects is non-classical, that is, in terms of not exclusively numerical or categorical features, but simultaneously appear both kinds of values feature and also, sometimes, it is necessary to employ a special symbol to denote the absence of values (missing values). That is, we could be dealing with mixed incomplete description of objects. In other words, we are talking about the process of certain amount of mixed incomplete data. Why and where was born the logical combinatorial approach to PR? PR problems with mixed incomplete description of objects frequently appear in Soft Sciences. For example: the description of geological zones for the establishment of some prognosis about mineral resources [1, 4]. For many practical problems geophysicists need to take into account not only measures like magnetic fields, gravimetric anomaly of Bouguer, intensity of the air magnetic field, an other numerical variables, but also features like a genetic series of soil, the presence or absence of faults, level of development of the soil, and others. You can say: well, let us to process all categorical features and make one decision, then process all numerical variables and make a second one. Finally, you can arrive at a conclusion over the partial decisions made before. That is, as any problem of collective decision-making. But the problem is, as you can see obviously, that non-categorical variable is analyzed together some numerical variable. For the geophysicists this is a big problem. It is not possible to take a based solution in these conditions. Other interesting example appears in Medicine. The knowledge representation in general, but in particular medical knowledge involved in the process of medical diagnosis or prognosis of health is a non-trivial problem because of its inherent subjectivity [5-13]. Normally, a physician evaluates signs and symptoms in the patient to establish a preventive diagnosis or prognosis of health. Signs are usually numbers. For example temperature, blood pressure, age, number of children, and so on. But there are signs that you cannot put in correspondence with a number, but with a code. For example: pallor, sweating, trembling, and others. Obviously, symptoms are subjective. What the patient is feeling depends on the person; nevertheless these are very important data for the physician. It is not too difficult to see the same problem as described above in the case of Geosciences. In both of these cases objects (zones, patients or diseases) are just n-ary tuples, elements of simple Cartesian product without any algebraic, logical or topological properties assumed on this space. How then, do we select in these cases the most informative features, classify a new object given a (training) sample of each classes of the problem or find the relationships between all objects based on a certain measure of similarity? This is the main problem resolved by LCPR. This approach is a necessary methodological answer to the fact that in many feature selection and classification problems, which frequently appear in Soft Sciences, the descriptions of objects are mixed and incomplete data. We said that this approach originated in the Soviet Union, with the publication of [see 8 Other references O1] in 1966. The principal task of above paper was to find a solution to a real problem: the prognosis of mineral resources. For the solution of this problem it was necessary to take into account, not only the geophysical measure of different phenomena, but also the type of soil, the present or absence of faults, and other features, which have only a qualitative character. Actually, in the mentioned paper the authors consider that all of the features are Boolean and the problem was solved in a qualitative framework in a very simple way. Later were appearing other works in which was considered features of different kinds of nature, not only Boolean and not only qualitative features [14-21]. Any way, it is important to underline that this approach originated because the practical necessity to modeling real problems in a more adequate form. Two important reviews through selected works related on LCPR you can find in [22,23]. The first textbook in this matter was appearing in Mexico last year [24]. 2. COMPOSITION In this moment 5 persons with the following qualification are in the official staff of our Group: Dr. José Ruiz-Shulcloper (leader); Dr. Manuel Lazo-Cortés; Dr. Eduardo AlbaCabrera; MS. Ramón Pico-Peña; MS. Ignacio Sánchez-Gutierrez. There are also other professors and researchers working for other universities and researches centers, in Cuba and in Mexico, which participate in almost all researches of the Group: MS. Aurora Pons-Porrata; MS. Fernando Vázquez-Mesa; MS. Damaris Pascual-González; MS. Roxana Danger-Mercaderes; MS. Aimé Alvarez-Roca at the Oriente University, Cuba; Dr. Martha R. Ortiz-Posada at the Metropolitan Autonomy University, Mexico; MS. José F. Martínez-Trinidad; MS. Jesús A. Carrasco-Ochoa; MS. Guillermo SánchezDíaz at the Computer Researches Center, IPN, Mexico; MS. Verónica Jiménez-Jacinto; Lic. Saturnino J. Morales-Escobar; María E. Guevara-Cruz at the Technological University of Mexico. Cuban and Mexican undergraduate and postgraduate students are working too in this area. It is necessary underline that our group has a useful professor and researcher interchange with several universities and researches centers in Cuba, Mexico, Portugal, Spain and USA. These relationships allow making more efficient our scientific work. 3. TEMATICS OF WORK Starting from the concepts of mixed incomplete data and similarity function our Group had aboard all related classification and recognition problems. In feature selection problems the principal tool employed was Testor Theory. This is a branch of Mathematical Logic that began in the Soviet Union at the end of the 50’s. I. A. Cheguis and S. V. Yablonskii [O2] were the first researchers that developed this theory. Their works were motivated by the problem of fault detection in logical schemes, particularly applied to computer logical circuits. To find a fault in complicated logical circuits is a very hard task; hence, the natural idea of attempting to simplify this. In the mentioned paper, authors considered a table of Boolean functions with n Boolean variables. These functions model all possible faults of a circuit: each function describes the behavior of the circuit when a certain fault occurs. Values of function arguments are interpreted as circuit states, and the values the function takes as the circuit response to the states. For Cheguis and Yablonskii, a testor is a collection of values of arguments (function inputs), such that, the respective responses (function outputs) for any pair of functions are different pair wise. In other words a testor is a set of circuit states that allows identifying the registered fault among all possible faults. The smaller the size of this set, the more useful the testor is. In the middle of the 60’s, Y. I. Zhuravlev adapted the testor concept to PR [O1]. It is worth mentioning that although a fault detection problem can be interpreted as a PR problem, Yablonskii did not do so. Suppose that a given training sample, in a framework of a supervised classification problem, has a (training) matrix representation MI, that is, object descriptions are stored in a matrix with as many columns as features, as many rows as objects in the sample, and they are split in groups corresponding with their respective classes. If the complete set of features R allows us to distinguish between objects (rows of MI) from different classes, then R is a testor. Furthermore, any non-empty feature subset of R, that satisfies this property, is a testor (in the Zhuravlev’s way). From the set of all testors of a matrix MI, there are some testors, which are irreducible. That is: if any feature is eliminated from them, then they stop being testors. That means, they confuse objects belonging to different classes. These testors we are called typical testors. Initially Zhuravlev, in order to simplify, supposed that MI had objects of only 2 disjoint classes, and features are Boolean, but of course all this is valid for more than two classes and the extension to non-Boolean feature is not difficult. Obviously, although the capability for modeling in a more adequate form increases in this last case, also the complexity of algorithms increase with non-Boolean features. We can emphasize that, besides Zhuravlev’s concept (and the natural extension previously exposed), in literature there are many other different testor concepts [O3, O4, 25-37-12]. These concepts were introduced taking into account different characteristics of real problems, which often appear in soft sciences (fuzzy classes, comparison criteria of similarity and dissimilarity simultaneously, real comparison criteria, similarity function different than feature wise total coincidence, etc). These extensions have the pretension to reach a more adequate model of the reality and this allows reaching better practical results. Our Group introduced many of these new concepts. In this field there are three principal tasks: adequate testor concept for the real world problems; efficient algorithms for the calculus of all typical (also other kinds) testor and the study of sensitivity of the set of typical testor in presence of changes in the training matrix; and all of them because the introduction of these concepts and tools in real world problems. Our Group has worked in all of these problems and has obtained a body of new results in each one. As it was mentioned before, different concepts of testor were introduced in order to give a better answer to different real world situations. Several algorithms for the calculate all typical testors [38-47] were implemented and employed in the resolution of real problems like the determination of the influence of the certain social-economic features in the analysis of young delinquency [14] or the informational weight of the geological and geophysical characteristics in the Gasopetroliferous forecast in Cuban ophiolitic association [4], the relevance of symptoms and signs in the evaluation of patients with clefts of the lip and palate [6,9,10,12], and others applications in Geosciences [2,4,48,49] or Medicine [13,50] and others [51,52]. The problem of the way to determinate the relevance of features and objects also was studied by the Group and interesting results you can find in [43,53-55]. In supervised classification problems the Group has been developed so called partial precedence algorithms. This is a series of families of parametric algorithms based on the partial precedence principle meaning that physicians, and other natural scientists on the base of partial similarities of the objects establish the similarity object wise in real world problems. An important characteristic of these algorithms is the assumption that the representation space of objects has not any algebraical, topological or logical structure. That is, they are only Cartesian products. These families were born in Soviet Union [O1,O5,O6] and our Group continues their development. As our philosophy is to make the mathematical and computer tools needed by the practical problems, we developed several new extensions of these families of partial precedence algorithms [10,21,56-59]. The classical voting algorithm introduced by Zhuravlev was extended to different real world situations and were introduced new similarity functions as well as fuzzy set theory concepts in order to adequately model problems in Geosciences. Our Group developed other two families of partial precedence algorithms. The KORA-Ω algorithm [57] is an extension of the known old and very used in Geosciences algorithm KORA-3 [O5]. This algorithm, was introduced by Bongard in 1963, was devised for the solution of supervised classification problems in Geosciences. It works with two disjoint classes and objects described in terms of Boolean variables without missing values. The underlying idea in the initial method is to classify new objects based on a training sample of objects through the verification of fulfillment of some complex properties. These properties are a set of three values (complex feature) corresponding to three features, which appear often enough in one of the classes and do not appear enough in the other one [O7]. Fuzzy KORA-Ω allows us to solve supervised classification problems with many classes (hard −disjoint or not− or fuzzy), with any kind of features and different types of similarity measures. In this model, complex properties could be of any length grater or equal than one. These complex properties could be of different representation levels. Also we developed CR+ algorithm. It is a family of algorithms employing the concept of representative set. This idea was introduced by Baskakova and Zhuravlev [O6], based on the concept of partial precedence, they introduced the idea of rating information in favor of (positive representative set) and against (negative representative set) the ownership of the objects to the classes, as well as, considering that the parameters used for the classification should be associated to each class. We extended the model introducing a more general concept of representative set, which could be applied in conditions less restrictive, but that conserves like a particular case the original concept. Also we did some modifications to the algorithm in order to incorporate the new conditions and concepts, especially the case in which the classes are fuzzy sets, whereby we introduce a function, in this case lineal, for assigning belongings to new objects from the total evidence in favor of and against the ownership of the object to the class. In both last families we also developed several heuristics for the efficient estimation of some of parameters of each one, like support sets, similarity thresholds, informational weights of objects and features among others. All these algorithms were employed in several real practical researches in Medicine, Geosciences and others soft sciences [2,4,6,7,10,48-51]. Our Group also using both the partial precedence principle and Cartesian product representation space assumptions has faced the unsupervised classification problems. The results in this area are distributed in free unsupervised classification problems, [6064] restricted unsupervised classification problems [65,66], and conceptual clustering problems [67-72]. In each of these three areas we have been worked in the development of new concepts, which better allow modeling real world situations; algorithms, which realizes these new ideas and allow the introduction of them in the practice; and sensitivity problems, which improve the computational behavior in problems with changes on the initial information. In free clustering problems, we most employ clustering criteria as β 0-connected components, β 0-compact sets, β0-strictly compact sets; and β0-complete maximal sets. We found the set theoretical relationship between all of these criteria, both in hard and fuzzy cases. These relationships allow making a better analysis of different data sets in real problems. Also were developed efficient algorithms for calculating all clusters in each of criteria mentioned above. In restricted clustering problems in the framework of LCPR were obtained some interesting results like an extension of the c-means algorithm for the case of mixed incomplete data and non-necessarily-distance functions and a series of theorems that estimate all connected components of the sample. The conceptual clustering problems introduced by Michalski and his followers, was aboard since semantic analysis to the development of fuzzy conceptual tools for the solution of these problems. Connected very hardly with these problems our Group has continued also in the line of symbolic object theory introduced by Diday. In this last one, were obtained as new formulations of the concept of symbolic objects as new algorithms allow us the structuralization of conceptual spaces [73,74]. An important improve in these areas was the introduction of the Fuzzy Set Theory. All of above-mentioned PR problems were aboard also in fuzzy cases. That is, given the assumption that the classes were fuzzy sets, or among the feature sets there were some fuzzy or linguistic variables, or among the similarity comparison criteria of features were some real functions; and others situations carried on to fuzzy modeling as one of the best way for object, feature and its relationship representations. In all these cases were introduced new concepts, and algorithms [75-80]. An important practical problem is created when, in feature selection or any classification problems, there appear modifications on the initial data set. This fact could change a solution obtained earlier. In many cases the cost of the re-calculation of these results is very high and have sense to find heuristics allow obtaining new results to these problems. This is called sensitivity PR problem. In this sense our Group had obtained some hopefully results. Actually in this area there are many hard problems yet not resolved. As we mentioned above, our objective is to develop mathematical and computer tools for resolution of real world problems, especially but not exclusively in soft science areas. Hence, we developed several systems, and software libraries allow introducing these mathematical tools [3,24,36,38-55,57,60-65,67,69,73,7481-84]. Among of these programs PROGNOSIS (a Institute of Cybernetics, Mathematics and Physics product) for the solution of problems in Geosciences and more recently CLASITEX+ (a Software Pro International product) for the discovering of main theme in textual documents, are two of the relevant systems in which our group had participated. Recently the Group starts to aboard the problem of enlargement of amount of data also in the framework of LCPR assumptions. Hence, were enfaced two related problems: Data Mining and Data Fusion. That is, we star to develop the called Mixed Incomplete Data Mining, and Mixed Incomplete Data Fusion as a logical consequence of the development of the LCPR theory. In the first case were obtained three new clustering algorithms [81-83] allow dealing with large and very large mixed incomplete data set. Also some primary results were obtained in Text Mining [84]. At present we are stating the theoretical bases of the fusion of mixed data hardly connected with the researches of IRIS Laboratory of Dr. Abidi, the combining classifiers problems developed by Dr. Kittler, and the schemas for collective decision problems developed by Dr. Kuntcheva. In the following figure it is shown the relationships between these areas that look like different appearances, but we think they are a whole consistent entity. In some sense it show also the future researches, which will be developed by our Group. PATTERN RECOGNITION D A T A F U S I O N Logical Combinatorial Pattern Recognition P R A Mixed Incomplete Data Fusion C T I C Mixed Incomplete Data Mining E DATA MINING 4. PROJECTS In this moment the Group is dealing basically with the following research projects: Mathematical models and computer tools for the solution of PR problems appearing in Soft Sciences. Automatically classification of phyto-pathologic relevant microscopic mushroom based on exploratory data analysis and advance image processing techniques. Processing of huge amount of heterogeneous information: Mixed Incomplete Data Mining 5. FINANCES Our projects are financed through Science and Technical Agency, which belongs to Science and Technical Ministry. 6. MORE RELEVANT PAPERS 1. E.N. Cheremesina; José Ruiz-Shulcloper. Cuestiones metodológicas de la aplicación de modelos matemáticos de Reconocimiento de Patrones en zonas del conocimiento poco formalizadas. Revista Ciencias Matemáticas, vol. 13; No.2; pp. 93-108, Cuba, (1992). 2. César Alaminos, José‚ Ruiz Shulcloper, Ramón Pico y otros. Aplicación de los Modelos de Reconocimiento de Patrones Lógico-Combinatorios a la resolución de tareas geológicas. Memorias de GEOLOGIA'89. 3. José Ruiz-Shulcloper, R. Pico Peña et al. PROGNOSIS y sus aplicaciones a las Geociencias, Revista Ciencias Matemáticas, Vol. 14, No. 2-3, 1993. pp. 124-144. Also in: III Congreso Iberoamericano de Inteligencia Artificial. IBERAMIA'92 Proceedings Ed. Limusa, México. pp. 561-586, (1992). 4. Julio Gómez-Herrera; Rodríguez Morán,O.; Valladares Amaro, S.; Ruiz Shulcloper,J.; Pico Peña, R. y otros. Prognostic of Gas-oil deposits in the Cuban ophiological association, applying mathematical modeling. Geophysics International 33, 3, pp. 447-467 (1995). 5. Martha R. Ortiz Posadas, José Francisco Martínez-Trinidad, José Ruiz-Shulcloper. A new approach to differential diagnosis of diseases, International Journal of Biomedical Computing, 40, pp. 179-185 (1996). 6. Martha R. Ortiz Posadas, y M. Lazo Cortés, Evaluación de la rehabilitación de pacientes con fisuras de paladar utilizando técnicas de reconocimiento de patrones, II Taller Iberoamericano de Reconocimiento de Patrones. Conferencia Internacional CIMAF'97. La Habana. Memorias (ISBN 968-29-9892-1), pp. 231-236, 1997. 7. Martha R. Ortiz Posadas, J. Maya Behart y M. Lazo Cortés. Evaluación de la cirugía de labio y paladar hendido con el enfoque lógico combinatorio de la teoría de reconocimiento de patrones. Revista Brasileira de Bioengenharia. Caderno de Engenharia Biomedica, v. 14, n. 1, pp. 7-22, (1998). 8. Martha R. Ortiz Posadas, L. Vega Alvarado, V. Jiménez Jacinto y M. Lazo Cortés. El concepto de analogía en medicina. Una función de semejanza para pacientes con fisura de paladar. III Taller Iberoamericano de Reconocimiento de Patrones. México. Memorias (ISBN 970-18-1081-3), pp. 247-256, 1998. 9. Martha R. Ortiz Posadas, L. Vega Alvarado, V. Jiménez Jacinto y M. Lazo Cortés. Una herramienta para evaluar la calidad del servicio que ofrece una clínica multidisciplinaria de labio-paladar hendido. I Congreso Latinoamericano de Ingeniería Biomédica. Mazatlán, México. pp. 796-799, (1998). 10. Martha R. Ortiz Posadas, L. Vega Alvarado, V. Jiménez Jacinto, M. Lazo Cortés y J. Maya Behart. Pronóstico de la rehabilitación de pacientes con fisuras de paladar usando un algoritmo de precedencia parcial. IV Simposio Iberoamericano de Reconocimiento de Patrones. Conferencia Internacional CIMAF’99. La Habana. Memorias (ISBN 970-18-2386-9), pp. 411-418, (1999). 11. Martha R. Ortiz-Posadas. Elaboration and application of a new pattern recognition methodology in Medicine. Ph. D. Thesis, Universidad Autónoma Metropolitana, Iztapalapa Campus, Mexico City (1999). 12. Martha R. Ortiz-Posadas. Prognosis and evaluation of cleft palate patients rehabilitation using pattern recognition techniques. World congress on medical physiscs and Biomedical Engineering 35, 1, 500. Niza, France (1997). 13. Rafael Sánchez Aroche and M. Lazo Cortés. Automatic auditory brainstem response (ABR) classification using the Convex method for an optimal feature set. International Journal of Medical Informatics (accepted) 14. José Ruiz Shulcloper y Alberto Fuentes. Un modelo cibernético para el análisis de la delincuencia juvenil. Revista Ciencias Matemáticas Vol. II(1) pp.141-153. (1981). Cuba 15. José Ruiz Shulcloper y Orlando Olivera. Algunas perspectivas de la aplicación de los modelos matemáticos al desarrollo del deporte en Cuba. Revista Control, Cibernética y Automatización, año XVI(1), enero-marzo pp. 3-7. Cuba. (1982). 16. José Ruiz Shulcloper. Elaboración de la información Biomédica por medio de modelos de reconocimiento de patrones. Revista Control, Cibernética y Automatización, año XVI(4), octubre-diciembre pp. 13-21. Cuba. (1982). 17. José Ruiz Shulcloper. Posibilidades del Reconocimiento de Patrones en el diagnóstico de enfermedades. Revista Control, Cibernética y Automatización, año XVI(1), enero-marzo pp. 10-13. Cuba. (1982). 18. Alberto Fuentes, José Ruiz Shulcloper y Luis Romero. Algoritmos de clasificación basados en el peso de los patrones. Revista Ciencias Matemáticas Vol. IV(1) pp.97121. Cuba. (1983). 19. José Ruiz Shulcloper. Problemas de clasificación de objetos y selección de variables en medios difusos: tres proyectos de tesis. Revista del Seminario de Enseñanza y Titulación. Año IV No. 21 pp. 41-68. México. (1988). 20. José Ruiz Shulcloper. Problemas actuales en la Teoría Matemática de Reconocimiento de Patrones. Memorias del Simposium Internacional de Computación, November 9-11, pp 1-42. México.(1994). 21. José Ruiz-Shucloper. Some questions about Pattern Recognition problems with nondisjoint classes. Editorial Academia, 1-32. (In Spanish). (1995). 22. José Francisco Martínez-Trinidad, Adolfo Guzmán-Arenas. The logical combinatorial approach to Pattern Recognition, an overview through selected works. (Accepted to Pattern Recognition). 23. Manuel Lazo-Cortés, José Ruiz-Shulcloper, Eduardo Alba-Cabrera. An overview of the evolution of the concept of testor in pattern recognition. (Accepted to Pattern Recognition). 24. José Ruiz-Shulcloper, A. Guzman-Arenas, and J. F. Martínez-Trinidad. Logical Combinatorial Approach to Pattern Recognition: I. Feature selection and supervised classification. Editorial Politécnica, Mexico (1999). Also will appear in the Editorial Fondo de Cultura Económica, Mexico, 150 pp. In Spanish. (2000). 25. José Ruiz-Shulcloper, M. Lazo Cortés. K-testores primos. Revista Ciencias Técnicas, Físicas y Matemáticas. (Cuba), vol. 9, pp. 17-55, (1991). 26. Manuel Lazo Cortés, and J. Ruiz Shulcloper. Determining the Feature Relevance in a Pattern Recognition Problem in Fuzzy Environments. Second European Congress on Intelligent Techniques and Soft Computing. EUFIT 94. Aachen, Germany. Proceedings, vol. 1, pp. 221-226, (1994). 27. Manuel Lazo Cortés. Una generalización del concepto de testor. Aportaciones Matemáticas. Serie Comunicaciones (IMATE-UNAM. México), No.14, pp. 283-288, (1994). 28. José Ruiz Shulcloper, Eduardo Alba Cabrera, et. al. Tópicos acerca de la Teoría de Testores. Serie Amarilla No 134 (investigaciones), pp. 1-6, CINVESTAV - IPN. México. (1994). 29. Eduardo Alba Cabrera, Nancy López Reyes, José Ruiz Shulcloper. Extensión del concepto de testor típico a partir de la función de analogía entre patrones. Algoritmos para su cálculo. Serie Amarilla No 134 (investigaciones), pp. 7-28, CINVESTAV IPN. México. (1994). 30. Manuel Lazo Cortés. Testores generalizados. Revista Integración. Vol. 14, No. 1, pp. 31-47, Colombia, (1996). 31. Eduardo Alba Cabrera. Un concepto de testor para cualquier función de analogía con imagen en un conjunto ordenado totalmente. Revista Integración, Vol. 14, No. 2, pp. 75-82, UIS, Colombia, (1996). 32. Salvador Godoy Calderón, Manuel Lazo Cortés. ∆-testores. A generalization of the concept of testor in fuzzy environments. Proceedings of the Second Iberoamerican Workshop on Pattern Recognition, March 24-28, Havana, Ed. IPN, México, pp. 95103, (In Spanish). (1997). 33. Manuel Lazo Cortés y J. Ruiz Shulcloper. Extensiones Difusas del Concepto de Testor. I Taller Iberoamericano de Reconocimiento de Patrones. Conferencia Internacional CIMAF'95. La Habana, 1995. Memorias, edit. Instituto Tecnológico de Toluca. México. pp. 181-193, (1997). 34. Eduardo Alba Cabrera, Manuel Lazo Cortés. Nueva generalización del concepto de testor para clases difusas. Resúmenes del III Taller Iberoamericano de Reconocimiento de Patrones. México, D.F., pp. 237-246, Ed. IPN-México. (1998). 35. Eduardo Alba Cabrera, Manuel Lazo Cortés. Una solución global para la utilización de los testores en problemas de Reconocimiento de Patrones. Resúmenes del III Taller Iberoamericano de Reconocimiento de Patrones. México, D.F., pp. 209-218, Ed. IPN-México. (1998). 36. Manuel Lazo-Cortes, Mercedes Douglas de la Peña, Teresita Quintana Gómez. Testor by class: an application to character recognition. III Taller Iberoamericano de Reconocimiento de Patrones. México. Memorias (ISBN 970-18-1081-3), pp. 229-236, (1998). 37. Manuel Lazo Cortés y M. Douglas de la Peña. Extensiones del concepto de testor por clase. IV Simposio Iberoamericano de Reconocimiento de Patrones. Conferencia Internacional CIMAF’99. La Habana. Memorias (ISBN 970-18-2386-9), pp. 209-218, (1999). 38. José Ruiz Shulcloper, Luis Aguila y Adrián Bravo. Algoritmos para el tratamiento automático de la información relativa a la descripción y clasificación de objetos y fenómenos. Revista Ciencias Físico-Técnicas y Matemáticas, No.2, pp. 139-149. (1983). 39. José Ruiz Shulcloper, Luis Aguila y Adrián Bravo. Algoritmo BT y TB para el cálculo de todos los test típicos. Revista Ciencias Matemáticas Vol. VI(2) pp.11-18. (1985). 40. José Ruiz-Shulcloper; Pico Peña, R. SELECTOR:sistema herramienta de selección de variables y objetos. Proceeding INFORMATICA'90, 19-25/2/1990. ICIMAF, pp 86-95. Editorial Academia. Cuba. (1992). 41. Manuel Lazo Cortés y E. E. Barreto Fiu. Testores difusos. Un algoritmo para determinar los testores difusos típicos de una matriz de aprendizaje. Reporte Técnico del CINVESTAV-IPN. Serie Amarilla No.134 pp. 29-42. México. (1994). 42. Manuel Lazo Cortés y E. E. Barreto Fiu, Algoritmo TDT para el cálculo de los testores difusos típicos de una matriz de aprendizaje, Reporte Técnico del CINVESTAV-IPN. Serie Amarilla No.134 pp. 43-51. México. (1994). 43. Manuel Lazo Cortés and J. Ruiz Shulcloper. Determining the feature relevance for non-classically described objects and a new algorithm to compute typical fuzzy testors. Pattern Recognition Letters 16 1259-1265, (1995). 44. Eduardo Alba Cabrera. Cálculo de todos los testores típicos para matrices kvalentes y función de semejanza de un umbral lε. Resúmenes del I Taller Iberoamericano de Reconocimiento de Patrones. La Habana Cuba, pp. 155-169, Ed. Tec. de Toluca, México., (1995). 45. Mario Farías Elinos, P. Rayón Villela y M. Lazo Cortés. Programación paralela de un algoritmo para el cálculo de testores con PVM. II Taller Iberoamericano de Reconocimiento de Patrones. Conferencia Internacional CIMAF'97. La Habana. Memorias (ISBN 968-29-9892-1), pp. 149-156, (1997). 46. Guillermo Sánchez Díaz, M. Lazo Cortés y J. García Fernández. Modelos algorítmicos paralelos y distribuidos para el cálculo de testores típicos. II Taller Iberoamericano de Reconocimiento de Patrones. Conferencia Internacional CIMAF'97. La HabanaMemorias (ISBN 968-29-9892-1), pp. 135-140, (1997). 47. Guillermo Sánchez Díaz, M. Lazo Cortés y O. Fuentes Chávez. Algoritmo genético para calcular testores típicos de costo mínimo. IV Simposio Iberoamericano de Reconocimiento de Patrones. La Habana. Memorias (ISBN 970-18-2386-9), pp. 207213, (1999). 48. José Ruiz-Shulcloper; Pico Peña, R. et al. Modelación Matemática del pronóstico de magnitudes máximas sísmicas de los terremotos en la Región del Caribe. In Reconocimiento de elementos de estructuras espaciales. pp. 81-101. Editorial Academia. Cuba. (1992). 49. José Ruiz-Shulcloper; Pico Peña, R.; Alaminos Ibarría, C.; Manchado Martín, A. y Valdés Hernández, Ma. G. Modelación matemática del problema de discriminación de anomalías AGE perspectiva para rocas fosfóricas de génesis sedimentaria. Revista Ciencias Matemáticas. Vol. 13, No. 2, (1993). 50. Sergio López Pérez, M. Lazo Cortés y H. M. Estrada García. Electrodiagnóstico médico utilizando las herramientas de reconocimiento de patrones. II Taller Iberoamericano de Reconocimiento de Patrones. Conferencia Internacional CIMAF'97. La Habana. Memorias (ISBN 968-29-9892-1), pp. 237-244, (1997). 51. Ramón Pico Peña, G. Almagro, J. Mena. Aplicación de técnicas de Reconocimiento de Patrones con enfoque lógico-combinatorio en la clasificación de un complejo de géneros de hongos patógenos de la Caña de Azúcar. Memorias del III Taller Iberoamericano de Reconocimiento de Patrones. México. Ed IPN, México. ISBN 970-18-1081-3. (1998). 52. Manuel Lazo-Cortes, Mercedes Douglas de la Peña, Teresita Quintana Gómez. Testor by class: an application to character recognition. III Taller Iberoamericano de Reconocimiento de Patrones. México. Memorias (ISBN 970-18-1081-3), pp. 229-236, (1998). 53. Manuel Lazo Cortés y J. Ruiz Shulcloper. Evaluación de la relevancia de los rasgos, a partir del concepto de testor, en problemas de clasificación. I Taller Iberoamericano de Reconocimiento de Patrones. Conferencia Internacional CIMAF'95. La Habana, 1995. Memorias, edit. Instituto Tecnológico de Toluca. México. pp. 195-204, (1997). 54. Manuel Lazo Cortés, and J. Ruiz Shulcloper. Determining the Feature Relevance in a Pattern Recognition Problem in Fuzzy Environments. Second European Congress on Intelligent Techniques and Soft Computing. EUFIT 94. Aachen, Germany. Proceedings, vol. 1, pp. 221-226, (1994). 55. Manuel Lazo Cortés, G. Sánchez Díaz y T. Quintana Gómez. Evaluación de la relevancia de los rasgos en un problema de clasificación supervisada a partir de todos los testores. IV Simposio Iberoamericano de Reconocimiento de Patrones. Conferencia Internacional CIMAF’99. La Habana. Memorias (ISBN 970-18-2386-9), pp. 215-222, (1999). 56. José Ruiz Shulcloper. Modelo de Algoritmos de Reconocimiento con Aprendizaje Parcial. Memorias del III Congreso Iberoamericano de Inteligencia Artificial, IBERAMIA’92. February 17-22, La Habana, Cuba pp. 541-559. Editorial Limusa, México. (1992). 57. Jesús Ariel Carrasco Ochoa, José Ruiz-Shulcloper, Lucía de la Vega Doria. Fuzzy KORA-Ω algorithm. Proceedings of the Sixth European Congress on Intelligent Techniques and Soft Computing, EUFIT’98, pp.1190-1194, Aachen, Germany (1998). 58. José Ruiz-Shulcloper and M. Lazo-Cortés. Mathematical Algorithms for the Supervised Classification Based on Fuzzy Partial Precedence. Mathematical and Computer Modelling 29, 4, pp. 111-119 (1999). 59. José Ruiz-Shulcloper y M. Lazo Cortés. Herramientas para la clasificación supervisada basadas en analogías parciales. Revista Mexicana de Ingeniería Biomédica (ISSN 0188-9532) (México), vol. 16, No. 2, pp. 75-93, (1995). 60. José Fco. Martínez Trinidad, J. Ruiz Shulcloper y M. Lazo Cortés. Criterios agrupacionales no clásicos para la estructuración de universos. Simposium Internacional de Computación, México. Memorias (ISBN:968-29-9545-0). pp. 251270, (1996). 61. Leonardo Alvarez; Pico, R.; Cotilla,M. Clasificación no supervisada por métodos lógico-combinatorios en problemas de zonación Sísmica. Reporte de Investigación. CEMAFIT. ICIMAF. (1995). 62. Ramón Pico Peña. Determinación del umbral de semejanza β0 para los algoritmos lógico-combinatorios, mediante el dendrograma de un algoritmo jerárquico. Memorias del IV Simposio Ibero-americano de Reconocimiento de Patrones (SIARP’99). Cuba. Ed IPN, México. ISBN 970-18-2386-9. (1999). 63. Ramón Pico Peña; Alvarez,L. y Cotilla,M. Zonación Sísmica de la Isla de Cuba mediante algoritmos de clasificación lógico-combinatorios. Memorias del II Taller Iberoamericano de Reconocimiento de Patrones. Ed. IPN, México. ISBN 968-299892-1. (1997). 64. Ramón Pico Peña. Determinación del umbral de semejanza β0 para los algoritmos lógico-combinatorios, mediante el dendrograma de un algoritmo jerárquico. Memorias del IV Simposio Ibero-americano de Reconocimiento de Patrones. Cuba. Ed IPN, México. ISBN 970-18-2386-9. (1999). 65. J.R. García-Serrano, José Francisco Martínez-Trinidad. Extension to C-means algorithm for the use of similarity functions. Proc. 3rdEuropean Conference on Principles and Practice of Knowledge Discovery in Databases, pp. 354-359, Prague, (1999). 66. R. Reyes-González J. Ruiz-Shulcloper. Un algoritmo de estructuración restringida de espacios. Memorias del IV Simposio Iberoamericano de Reconocimiento de Patrones, March 22-27, 1999, Ciudad de la Habana. pp. 267-278. Edited by José Ruiz Shulcloper, José Fco. Martínez Trinidad, Jesús Ariel Carrasco Ochoa, Guillermo Sánchez Díaz, Stalin Muñoz Gutiérrez, Walterio Mayol Cuevas, pp. 683. Editorial Politécnico. México. (1999). 67. José Francisco Martínez-Trinidad, José Ruiz-Shulcloper. LC-conceptual algorithm: characterization using typical testors by class. Proceedings of the Seventh European Congress on Intelligent Techniques and Soft Computing, EUFIT’99, on CD, Section FS C, Aachen, Germany (1999). 68. José Francisco Martínez Trinidad, José Ruiz Shulcloper, Aurora Pons Porrata. Algoritmo LC-Conceptual: Una mejora de la etapa intencional utilizando reglas de generalización. Memorias del Taller Nacional de Inteligencia Artificial TANIA’99, ISBN 970-18-3554-9, México, D.F. pp. 179-188, 1999. Also in: Informe Técnico; Serie Roja No. 70, pp. 1-10, Noviembre ISBN 970-18-3914-5, México. (1999). 69. José Fco. Martínez; J. Ruiz Shulcloper. Algoritmo LC-conceptual para el agrupamiento de objetos. Simposium Internacional de Computación, CIC’97, November 12-14, pp. 411-418, México. Also in: Informe Técnico Serie Roja, No. 27; pp. 1-11, Junio 1998. ISBN 970-18-1551-3, (1997). 70. José Fco. Martínez, J. Ruiz-Shulcloper. Un modelo de estructuración conceptual. Memorias del II Taller Iberoamericano de Reconocimiento de Patrones, March 2428, 1997, C. de la Habana. pp. 113-123. Edited by Adolfo Guzmán Arenas, José Ruiz Shulcloper, Humberto Sossa Azuela, Manuel Lazo Cortés, J. Luis Díaz de León Santiago. Instituto Politécnico Nacional, México. Informe Técnico Serie Roja, No. 29; pp. 1-8. ISBN 970-18-1553-X, (1998). 71. José Fco. Martínez; J. Ruiz Shulcloper. Estructuración conceptual de espacios I: Agrupamientos semánticos difusos. Informe Técnico Serie azul, No. 4, August, Editorial Politécnico, Mexico. (1996) 72. José Fco. Martínez Trinidad, J. Ruiz Shulcloper y M. Lazo Cortés. Criterios agrupacionales no clásicos para la estructuración de universos. Simposium Internacional de Computación, México, 1996. Memorias (ISBN:968-29-9545-0). pp. 251-270. Informe Técnico Serie Roja, No. 28, June, Editorial Politécnico; Mexico. (1998). 73. José Ruiz-Shulcloper, M. Chac, J.F. Martínez. Data analysis between sets of objects. Proceedings of the 8th International Conference on Systems Research, Informatics and Cybernetics, August, Baden-Baden, Germany. Edited by G. Lasker. Advances in Artificial Intelligence and Engineering Cybernetics, Vol. III, pp. 81-85, (1996). 74. José Ruiz-Shulcloper, M. Chac, J.F. Martínez. Cluster Analysis in Conceptual Spaces. Proceedings of the 9th International Conference on Systems Research, Informatics and Cybernetics, August, Baden-Baden, Germany. Edited by G. Lasker. Advances in Artificial Intelligence and Engineering Cybernetics, Vol. IV, pp., (1997). 75. José Ruiz-Shulcloper and J. J. Montellano-Ballesteros. A new model of fuzzy clustering algorithms. Proceedings of the European Congress on Intelligent Techniques and Soft Computing, EUFIT’95, pp. 1484-1488, Aachen, Germany (1995). 76. José Francisco Martínez, José Ruiz-Shulcloper. Fuzzy semantic clustering. Proceedings of the Fourth European Congress on Intelligent Techniques and Soft Computing, EUFIT’96, pp. 1397-14101, Aachen, Germany (1996). 77. José Francisco Martínez-Trinidad, José Ruiz-Shulcloper. Fuzzy Conceptual Clustering, Proceedings of the Fifth European Congress on Intelligent Techniques and Soft Computing, EUFIT’97, pp. 1852-1857, Aachen, Germany (1997). 78. José Francisco Martínez-Trinidad, José Ruiz-Shulcloper. Fuzzy LC conceptual algorithm. Proceedings of the Sixth European Congress on Intelligent Techniques and Soft Computing, EUFIT’98, pp. 20-24, Aachen, Germany (1998). 79. José Francisco Martínez Trinidad, José Ruiz-Shulcloper, Manuel Lazo Cortés. Structuralization of Universes. Fuzzy Sets and Systems, Vol. 112, No. 3, pp. 485-500, (2000). 80. José Francisco Martínez Trinidad, José Ruiz-Shulcloper. Fuzzy Clustering of Semantic Spaces. (Accepted to Pattern Recognition). 81. José Ruiz Shulcloper, Eduardo Alba Cabrera. Mixed Incomplete Data (MID) MINING for Spatial Databases. CD Memorias de GEOINFO 2000 ISSN 1028-8961. (2000). 82. José Ruiz Shulcloper, Eduardo Alba Cabrera, Guillermo Sánchez Díaz. Discovering β0-Density Connected Components from Large Mixed Incomplete Data Sets. CD Memorias de GEOINFO 2000 ISSN 1028-8961. (2000). 83. José Ruiz Shulcloper, Eduardo Alba Cabrera, Guillermo Sánchez Díaz. DGLC: a Density-Based Global Logical Combinatorial Clustering Algorithm for Large Mixed Incomplete Data. Accepted to IGARSS 2000 Honnolulu. (2000). 84. José Fco. Martínez Trinidad, Beatriz Beltrán Martínez, Adolfo Guzmán Arenas, José Ruiz Shulcloper. CLASITEX+: A tool for Knowledge Discovery from Texts. In: Principles of Data Mining and Knowledge Discovery. Second European Symposium, PKDD'98, Nantes, France, Sept. 1998. Edited by Jan M. Zytkow, Mohamed Quafafou. Lecture Notes in Artificial Intelligence, Vol. 1510. pp. 459-467. Informe Técnico Serie Verde, No. 19; pp. 1-9, Mayo 1999. ISBN 970-18-3083-0. (1998). 7. ASSOCIATIONS AND GROUPS Our Group belongs to the: a) Cuban Association for Pattern Recognition. b) Cuban Association of Mathematics and Computation. 8. OTHER REFERENCES There are many other important literature in Russian and in other languages that we are not cited because the restriction on the length of the paper. We are sorry with those authors for the omission. O1.A.N. Dmitriev, Yu. I. Zhuravlev, and F.P Krendelev. About the mathematical principles of objects and phenomena classification. Sbornik Diskrtnii Analisis 7, pp. 3-15. Novosibirsk . (In Russian). (1966). O2.I.A. Cheguis, and S.V. Yablonskii. Logical Methods for controlling electrical systems. Trudy Matematicheskava Instituta imeni V. A. Steklova LI, pp. 270-360. Moscow. (In Russian). (1958). O3.N. N. Aizenberg, and A. I. Tsipkin. Prime Tests. Doklady Akademii Nauk 201 (4), pp. 801-802. (In Russian). (1971). O4.R. S. Goldman: Problems of Fuzzy Testors Theory. Avtomatika i Telemejanika 10, pp. 146-153. (In Russian). (1980). O5.M.N. Bongard, et al. Solving geological problems using recognition programs. Sovietic Geology, No. 6, Moscow. (In Russian). (1963). O6.L.V. Baskakova, Yu.I. Zhuravlev. Model of algorithm of recognition with representative sets and support sets systems. Zhurnal Vichislitielnoi Matemati y Matematicheskoi Fisiki 21-5, 1264-75. (In Russian). (1981). O7. V. Keilis-Borok, A. Soloviev. Pattern Recognition in Earthquake Prediction. Workshop on Non-Linear Dynamics and Earthquake Prediction, ICTP, Trieste, Italy, pp. 1-14, (1991).