Seminaris de Secció d'IA

 


   

 Dimecres 17/11/1999

Marta Gatius / LSI -UPC

Títol: Adapting general linguistic knowledge to applications in order to obtain friendly and efficient NL interfaces

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum:  This work proposes a mechanism to adapt general linguistic knowledge to applications in order to obtain automatically application-restricted NL interfaces. The knowledge involved in the process is represented in separate data structures: a conceptual ontology, a linguistic ontology, a general lexicon and a set of control rules. The general and application-dependent knowledge relevant to the communication tasks is represented in a conceptual ontology. The general linguistic information is represented in a linguistic ontology. The core of our approach consists of describing the application elements (both entities and operations) relevant to the communicative tasks in terms of the existing conceptual ontology. Lexical coverage of these elements has to be provided as well. A general basic set of control rules will then relate the application specifications to the linguistic ontology in order to obtain application-restricted NL interfaces. This set of rules, implemented in the Production Rules Environment (PRE), has been applied to an expert system in law to obtain an application-restricted task-guided interface.

Paraules Clau: Adapting general linguistic knowledge, conceptual ontology.

Documentació: Text

 

 Dimecres 24/11/1999

Luis Molina / LSI - UPC

Títol: Uso de Técnicas de Aprendizaje Automático en la obtención de mejor ganado vacuno

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum: Ante la gran demanda de alimentos para el próximo siglo, la genética se está convirtiendo en una ciencia con un papel relevante para poder combatir el hambre. Desde 1970, fue creado en Brasil, el Programa de Mejoramiento Genético de la Raza Vacuna Nelore, en el cual ganaderos, universidades y asociaciones unen sus esfuerzos para obtener mejores animales, a partir de sus datos genéticos. Durante este tiempo se ha colectado electrónicamente la información genética de más de 130.000 animales de 60 haciendas ganaderas. Usando esta información, en está charla, se mostrarán los resultados de aplicar varios algoritmos de Aprendizaje Automático, para obtener algún conocimiento en relación a la obtención de mejores perímetros testiculares y de animales jóvenes con más peso. Los algoritmos C4.5 (árboles de decisión), CN2 y C4.5rules (reglas de producción), BKD (Redes Bayesianas) y Grobian (Rough Sets) fueron usados

Paraules Clau: Aprendizaje Automático, ganado bovino.

Documentació: Text

 

  Dimecres 1/12/1999

Ramon Sangüesa / LSI - UPC

Títol: Sistemes Recomanadors

Lloc: Vilanova.

Hora: de 10 a 11

Resum: 

Paraules Clau:

Documentació:

 

 Dimecres 15/12/1999

Lluís Belanche / LSI - UPC

Títol: Redes Neuronales Heterogéneas: Fundamentos y Aplicaciones

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum: Fuzzy heterogeneous networks are feed-forward neural network models composed of neurons of a general class whose inputs and weights are mixtures of continuous variables (crisp and/or fuzzy) with discrete quantities, also admitting missing data. These networks have net input functions based on similarity relations between the inputs to and the weights of a neuron. They thus accept heterogeneous --possibly missing-- inputs, and can be coupled with classical neurons in hybrid network architectures, trained by means of genetic algorithms or other evolutionary methods. The effectiveness of these networks is compared against more classical neural networks in the context of an investigation in the field of environmental sciences: the geochemical study of natural waters in the Arctic (Spitzbergen). Classification accuracy, the effect of working with crisp or fuzzy inputs, the use of traditional scalar product vs. similarity-based functions, and the presence of missing data, are studied.  

The results obtained show that, from these standpoints, fuzzy heterogeneous networks based on similarity perform better than classical models. This behaviour is consistent with previous results in other application domains.

Paraules Clau: fuzzy heterogeneous networks, heterogeneous inputs.

Documentació: Referéncia

 

 Dimecres 22/12/1999

Jordi Atserias / LSI - UPC

Títol: Using a diathesis model for Semantic Parsing

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum: I'll present the different techniques we have explored for the Semantic Parsing of unrestricted texts. The aim of the system is to obtain a case-role analysis of a sentence, in which the semantic roles of the verb are identified. Our approach is based on the identification of the resulting parse tree of the sentence as one of the models which describrs the different posible syntactic realization of the verb arguments. The approaches we have explored for recognizing the verb models are:

- Approximated tree patterning matched based on edit operation.

- Relaxation Labeling techniques, Based on the formalization of pattern-matching and pattern-combination as a Consistence Labelling Problem.

Paraules Clau: Natural Language Understanding, Semantic Parsing, Pattern Matching, Consistent Labeling Problem, Relaxation Labeling.

Documentació: Text1 Text2

 

 Dimecres 12/1/2000

Luigi Ceccaroni / LSI - UPC

Títol: Building ontologies

 Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum:  En el seminario hablaré de ontologías en general, de herramientas disponibles para construirlas (especialmente de las disponibles en el web) y brevemente de una ontología específica que estamos construyendo relativa al domínio del tratamiento de aguas residuales.

Paraules Clau: ontologies

Documentació: Presentació (PDF)

 

 Dimecres 19/1/2000

Títol:

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum: 

Paraules Clau:

Documentació:

 

 Dimecres 26/1/2000

Enrique Romero / LSI - UPC

Títol: Aproximación en espacios de Hilbert: Un método general secuencial y una implementación particular con redes neuronales

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum:  Function approximation is present in many different areas (Fourier series, polynomial approximation, statistics, signal processing, artificial intelligence, ...). The aim is very simple: expressing a function (more or less complex) as a (generally linear) combination of simple functions. Function approximation is a particular case of the more general approximation in Hilbert spaces (a Hilbert space is, essentially, a vector space with an inner product). In the theory, the approximations can consist of an infinite number of terms. In practical applications this is not possible. An attractive way to construct an approximation is, starting from scratch, adding terms one at a time to the partial approximations. This is the basis of the sequential approximation methods. The quality of the added terms is very important to achieve the desired approximation accuracy.

A new sequential method for approximating vectors in Hilbert spaces will be presented. The theoretical results obtained proof that, under reasonable conditions:

(1) the construction of the approximation is always possible

(2) any vector can be approximated by this approach

Besides, it seems that it should achieve the same accuracy that other existent sequential methods with fewer terms.

Also, a particular implementation using neural networks is presented, and some results discussed.

Paraules Clau:

Documentació: Número: LSI-00-13-R

 

 Dimecres 2/2/2000

Títol:

Lloc:

Hora:

Resum: 

Paraules Clau:

Documentació:

 

 Dimecres 9/2/2000

Títol:

Lloc:

Hora:

Resum: 

Paraules Clau:

Documentació:

 

 Dimecres 16/2/2000

Títol:

Lloc:

Hora:

Resum: 

Paraules Clau:

Documentació:

 

 Dimecres 23/2/2000

Joaquín Pina / LSI - UPC

Títol: Detección de puntos característicos en imágenes 3D

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum: 

Paraules Clau:

Documentació:

 

 Dimecres 1/3/2000

Lluís Màrquez / LSI - UPC

Títol: Boosting applied to Word Sense Disambiguation

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum: En aquesta presentacio parlare de l'aplicacio de tecniques de combinacio de classificadors al problema de desambiguar el sentit de les paraules (Word Sense Disambiguation). En concret presentare l'algoritme AdaBoost.MH (Schapire and Singer 98), i el resultats de l'experimentacio que hem portat a terme (Escudero, Marquez and Rigau 00) en el corpus etiquetat semanticament mes voluminos disponible actualment. Aquest corpus conte approx. 200.000 occurrencies de les 191 paraules (121 noms i 70 verbs) mes frequents i polisemiques de la llengua anglesa.

Paraules Clau: Natural Language Processing, Word Sense Disambiguation, Boosting algorithms, ensembles of classifiers, supervised learning.

Documentació:

 

 Dimecres 8/3/2000

Luigi Ceccaroni / LSI - UPC

Títol: Planning and Reactive Planning

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum: La charla será muy general sobre planificación y planificación reactiva.Veremos un poco de historia y el estado del arte de estas disciplinas. Acabaré con un pequeño ejemplo sencillo y concreto de planificación reactiva aplicada a procesos de depuración.

Paraules Clau:

Documentació:

 

 Dimecres 15/3/2000

Javier Béjar / LSI - UPC

Títol:

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum: 

Paraules Clau:

Documentació:

 

 Dimecres 22/3/2000

Lluís Vila / LSI - UPC

Títol:

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum: 

Paraules Clau:

Documentació:

 

 Dimecres 12/4/2000

Teresa Abad / LSI - UPC

Títol:

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum: 

Paraules Clau:

Documentació:

 Dimecres 26/4/2000

Núria Castell Arinyo / LSI - UPC

Títol: Projecte de diàleg i reconeixement de veu

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum: 

Paraules Clau:

Documentació:

 Dimecres 31/5/2000

Aida Valls / Universitat Rovira i Virgili

Títol:

Lloc: Biblioteca de LSI, Edifici C6-Campus Nord, 2a planta

Hora: de 10 a 11 (9:45 cafè i pastes)

Resum: 

Paraules Clau:

Documentació: