Učení neuronových sítí jako inverzní úloha
Se presenta una adaptación de la técnica Multiedit-Condensing (MC) recientemente aplicada a algunos problemas en Reconocimiento de Formas, a un problema de segmentación de imágenes que se encuadra dentro del marco general del Análisis de Imagen en Robótica. Se presenta una revisión de los fundamentos de dicha técnica y se justifica su uso para la solución del problema propuesto. Dicho problema consiste básicamente en la separación de las distintas zonas de interés en una imagen, según su color y...
We give an explicit criterion for unavoidability of word sets. We characterize extendible, finitely and infinitely as well, elements in them. We furnish a reasonable upper bound and an exponential lower bound on the maximum leghth of words in a reduced unavoidable set of a given cardinality.
L'apprentissage automatique à partir d'exemples consiste généralement à caractériser un ensemble d'objets dénotant un concept. Nous avons développé deux méthodes d'apprentissage symbolique, LEGAL et LEGAL-E, qui s'appuient sur le même modèle d'apprentissage, et utilisent une technique de généralisation descendante, basée sur la logique des propositions et sur la structure de treillis de Galois, pour produire un ensemble de descriptions structurées et ordonnées. Elles diffèrent dans leur approche...
Usually the problem of drift estimation for a diffusion process is considered under the hypothesis of ergodicity. It is less often considered under the hypothesis of null-recurrence, simply because there are fewer limit theorems and existing ones do not apply to the whole null-recurrent class. The aim of this paper is to provide some limit theorems for additive functionals and martingales of a general (ergodic or null) recurrent diffusion which would allow us to have a somewhat unified approach...
Similarity-based models were first used by Ruspini to give semantics to fuzzy logic. In these models, incomplete information is represented by an evidential set, i.e. a set of possible worlds that are compatible with the evidence, together with a fuzzy similarity relation} on the set of possible worlds that allows to describe the resemblance of arbitrary subsets of worlds to those belonging to the evidencial set. On the other hand, the question addressed by theory change formalisms is which kind...
Most of the techniques used in text classification are determined by the occurrences of the words (terms) appearing in the documents, combined with the user feedback over the documents retrieved. However, in our model, the most relevant terms will be selected from a previous fuzzy classification given by the genetic algorithm guided by the user feedback, but using techniques from Machine Learning. A feature selection process is carried out through a Genetic Algorithm in order to find the most discriminatory...