Scene Classification Using a Hybrid Generative/Discriminative Approach

We investigate whether dimensionality reduction using a latent generative model is beneficial for the task of weakly supervised scene classification. In detail, we are given a set of labeled images of scenes (for example, coast, forest, city, river, etc.), and our objective is to classify a new image into one of these categories. Our approach consists of first discovering latent ";topics"; using probabilistic Latent Semantic Analysis (pLSA), a generative model from the statistical text literature here applied to a bag of visual words representation for each image, and subsequently, training a multiway classifier on the topic distribution vector for each image. We compare this approach to that of representing each image by a bag of visual words vector directly and training a multiway classifier on these vectors. To this end, we introduce a novel vocabulary using dense color SIFT descriptors and then investigate the classification performance under changes in the size of the visual vocabulary, the number of latent topics learned, and the type of discriminative classifier used (k-nearest neighbor or SVM). We achieve superior classification performance to recent publications that have used a bag of visual word representation, in all cases, using the authors' own data sets and testing protocols. We also investigate the gain in adding spatial information. We show applications to image retrieval with relevance feedback and to scene classification in videos

Desenvolupament d'algorismes biològicament inspirats basats en Support Vector Machines

L'objectiu d'aquest projecte ha estat el desenvolupament d'algorismes biològicament inspirats per a l'olfacció artificial. Per a assolir-lo ens hem basat en el paradigma de les màquines amb suport vectorial. Hem construit algoritmes que imitaven els processos computacionals dels diferents sistemes que formen el sistema olfactiu dels insectes, especialment de la llagosta Schistocerca gregaria. Ens hem centrat en el lòbuls de les antenes, i en el cos fungiforme. El primer està considerat un dispositiu de codificació de les olors, que a partir de la resposta temporal dels receptors olfactius a les antenes genera un patró d'activació espaial i temporal. Quant al cos fungiforme es considera que la seva funció és la d'una memòria per als olors, així com un centre per a la integració multi-sensorial. El primer pas ha estat la construcció de models detallats dels dos sistemes. A continuació, hem utilitzat aquests models per a processar diferents tipus de senyals amb l'objectiu de abstraure els principis computacionals subjacents. Finalment, hem avaluat les capacitats d'aquests models abstractes, i els hem utilitzat per al processat de dades provinents de sensors de gasos. Els resultats mostren que el models abstractes tenen millor comportament front el soroll i més capacitat d'emmagatzematge de records que altres models més clàssics, com ara les memòries associatives de Hopfield o fins i tot en determinades circumstàncies que les mateixes Support Vector Machines.

Technologies de procédé et de contrôle pour réduire la teneur en sel du jambon sec et des saucissons

Dans certains pays européens, les produits carnés élaborés peuvent représenter près de 20% de la consommation journalière de sodium. De ce fait, les industries de la viande tentent de réduire la teneur en sel dans les produits carnés pour répondre, d’une part aux attentes des consommateurs et d’autre part aux demandes des autorités sanitaires. Le système Quick‐Dry‐Slice process (QDS®), couplé avec l’utilisation de sels substituant le chlorure de sodium (NaCl), a permis de fabriquer, avec succès, des saucisses fermentées à basse teneur en sel en réduisant le cycle de fabrication et sans ajout de NaCl supplémentaire. Les technologies de mesure en ligne non destructives, comme les rayons X et l’induction électromagnétique, permettent de classifier les jambons frais suivant leur teneur en gras, un paramètre crucial pour adapter la durée de l’étape de salaison. La technologie des rayons X peut aussi être utilisée pour estimer la quantité de sel incorporée pendant la salaison. L’information relative aux teneurs en sel et en gras est importante pour optimiser le processus d’élaboration du jambon sec en réduisant la variabilité de la teneur en sel entre les lots et dans un même lot, mais aussi pour réduire la teneur en sel du produit final. D’autres technologies comme la spectroscopie en proche infrarouge (NIRS) ou spectroscopie microondes sont aussi utiles pour contrôler le processus d’élaboration et pour caractériser et classifier les produits carnés élaborés, selon leur teneur en sel. La plupart de ces technologies peuvent être facilement appliquées en ligne dans l’industrie afin de contrôler le processus de fabrication et d’obtenir ainsi des produits carnés présentant les caractéristiques recherchées.

A genetic algorithm to ensemble feature selection

We are going to implement the "GA-SEFS" by Tsymbal and analyse experimentally its performance depending on the classifier algorithms used in the fitness function (NB, MNge, SMO). We are also going to study the effect of adding to the fitness function a measure to control complexity of the base classifiers.

Improving the organisation and commercialisation of small coffee and cocoa producers in the Northern Amazon Region of Ecuador

Coffee and cocoa represent the main sources of income for small farmers in the Northern Amazon Region of Ecuador. The provinces of Orellana and Sucumbios, as border areas, have benefited from investments made by many public and private institutions. Many of the projects carried out in the area have been aimed at energising the production of coffee and cocoa, strengthening the producers’ associations and providing commercialisation infrastructure. Improving the quality of life of this population threatened by poverty and high migration flows mainly from Colombia is a significant challenge. This paper presents research highlighting the importance of associative commercialisation to raising income from coffee and cocoa. The research draws on primary information obtained during field work, and from official information from the Ministry of Agriculture. The study presents an overview of current organisational structures, initiatives of associative commercialisation, stockpiling of infrastructure and ownership regimes, as well as estimates for both ‘robusta’ coffee and national cocoa production and income. The analysis of the main constraints presents different alternatives for the implementation of public land policies. These policies are aimed at mitigating the problems associated with the organisational structure of the producers, with processes of commercialisation and with environmental aspects, among others.

Validation procedures in radiological diagnostic models. Neural network and logistic regression

The objective of this paper is to compare the performance of twopredictive radiological models, logistic regression (LR) and neural network (NN), with five different resampling methods. One hundred and sixty-seven patients with proven calvarial lesions as the only known disease were enrolled. Clinical and CT data were used for LR and NN models. Both models were developed with cross validation, leave-one-out and three different bootstrap algorithms. The final results of each model were compared with error rate and the area under receiver operating characteristic curves (Az). The neural network obtained statistically higher Az than LR with cross validation. The remaining resampling validation methods did not reveal statistically significant differences between LR and NN rules. The neural network classifier performs better than the one based on logistic regression. This advantage is well detected by three-fold cross-validation, but remains unnoticed when leave-one-out or bootstrap algorithms are used.

Feature Extraction from Whole-Sky Ground-Based Images for Cloud-Type Recognition

Several features that can be extracted from digital images of the sky and that can be useful for cloud-type classification of such images are presented. Some features are statistical measurements of image texture, some are based on the Fourier transform of the image and, finally, others are computed from the image where cloudy pixels are distinguished from clear-sky pixels. The use of the most suitable features in an automatic classification algorithm is also shown and discussed. Both the features and the classifier are developed over images taken by two different camera devices, namely, a total sky imager (TSI) and a whole sky imager (WSC), which are placed in two different areas of the world (Toowoomba, Australia; and Girona, Spain, respectively). The performance of the classifier is assessed by comparing its image classification with an a priori classification carried out by visual inspection of more than 200 images from each camera. The index of agreement is 76% when five different sky conditions are considered: clear, low cumuliform clouds, stratiform clouds (overcast), cirriform clouds, and mottled clouds (altocumulus, cirrocumulus). Discussion on the future directions of this research is also presented, regarding both the use of other features and the use of other classification techniques

Counter-ion effect on surfactant-DNA gel particles as controlled DNA delivery systems

The ability to entrap drugs within vehicles and subsequently release them has led to new treatments for a number of diseases. Based on an associative phase separation and interfacial diffusion approach, we developed a way to prepare DNA gel particles without adding any kind of cross-linker or organic solvent. Among the various agents studied, cationic surfactants offered particularly efficient control for encapsulation and DNA release from these DNA gel particles. The driving force for this strong association is the electrostatic interaction between the two components, as induced by the entropic increase due to the release of the respective counter-ions. However, little is known about the influence of the respective counter-ions on this surfactant-DNA interaction. Here we examined the effect of different counter-ions on the formation and properties of the DNA gel particles by mixing DNA (either single- (ssDNA) or double-stranded (dsDNA)) with the single chain surfactant dodecyltrimethylammonium (DTA). In particular, we used as counter-ions of this surfactant the hydrogen sulfate and trifluoromethane sulfonate anions and the two halides, chloride and bromide. Effects on the morphology of the particles obtained, the encapsulation of DNA and its release, as well as the haemocompatibility of these particles, are presented, using the counter-ion structure and the DNA conformation as controlling parameters. Analysis of the data indicates that the degree of counter-ion dissociation from the surfactant micelles and the polar/hydrophobic character of the counter-ion are important parameters in the final properties of the particles. The stronger interaction with amphiphiles for ssDNA than for dsDNA suggests the important role of hydrophobic interactions in DNA.

Subclass problem-dependent design for error-correcting output codes

A common way to model multiclass classification problems is by means of Error-Correcting Output Codes (ECOCs). Given a multiclass problem, the ECOC technique designs a code word for each class, where each position of the code identifies the membership of the class for a given binary problem. A classification decision is obtained by assigning the label of the class with the closest code. One of the main requirements of the ECOC design is that the base classifier is capable of splitting each subgroup of classes from each binary problem. However, we cannot guarantee that a linear classifier model convex regions. Furthermore, nonlinear classifiers also fail to manage some type of surfaces. In this paper, we present a novel strategy to model multiclass classification problems using subclass information in the ECOC framework. Complex problems are solved by splitting the original set of classes into subclasses and embedding the binary problems in a problem-dependent ECOC design. Experimental results show that the proposed splitting procedure yields a better performance when the class overlap or the distribution of the training objects conceal the decision boundaries for the base classifier. The results are even more significant when one has a sufficiently large training size.

Participació i associacionisme de les dones a Vic

L’objectiu d’aquest treball és visibilitzar la participació de les dones a la ciutat de Vic, per tal de reconèixer el seu paper dins la societat. Aquest treball pretén doncs, ressaltar una presència visible i empoderada de les dones dins les associacions i, també, un canvi de mirada que permeti veure les dones com a protagonistes de canvis i transformacions. La divisió tradicional del treball va confinar la dona a l’espai privat, de manera que l’allunyà de la participació política, institucional i social de la vida. L’aportació del moviment associatiu femení ha estat clau, i la influència que ha exercit en les conquestes socials és evident. Com a producte final i per tal de donar evidència de la participació de les dones a la ciutat, adjunto un llibret on es pot trobar totes les associacions de dones que hi ha Vic, amb la seva fitxa tècnica pertinent i la mirada d’algunes de les dones que participen activament d’aquest associacionisme femení.

Maximum likelihood Linear Programming Data Fusion for Speaker Recognition

Biometric system performance can be improved by means of data fusion. Several kinds of information can be fused in order to obtain a more accurate classification (identification or verification) of an input sample. In this paper we present a method for computing the weights in a weighted sum fusion for score combinations, by means of a likelihood model. The maximum likelihood estimation is set as a linear programming problem. The scores are derived from a GMM classifier working on a different feature extractor. Our experimental results assesed the robustness of the system in front a changes on time (different sessions) and robustness in front a change of microphone. The improvements obtained were significantly better (error bars of two standard deviations) than a uniform weighted sum or a uniform weighted product or the best single classifier. The proposed method scales computationaly with the number of scores to be fussioned as the simplex method for linear programming.

A Robust Multiple Feature Approach To Endpoint Detection In Car Environment Based On Advanced Classifiers

In this paper we propose an endpoint detection system based on the use of several features extracted from each speech frame, followed by a robust classifier (i.e Adaboost and Bagging of decision trees, and a multilayer perceptron) and a finite state automata (FSA). We present results for four different classifiers. The FSA module consisted of a 4-state decision logic that filtered false alarms and false positives. We compare the use of four different classifiers in this task. The look ahead of the method that we propose was of 7 frames, which are the number of frames that maximized the accuracy of the system. The system was tested with real signals recorded inside a car, with signal to noise ratio that ranged from 6 dB to 30dB. Finally we present experimental results demonstrating that the system yields robust endpoint detection.

Empirical mode decomposition-based face recognition system

In this work we explore the multivariate empirical mode decomposition combined with a Neural Network classifier as technique for face recognition tasks. Images are simultaneously decomposed by means of EMD and then the distance between the modes of the image and the modes of the representative image of each class is calculated using three different distance measures. Then, a neural network is trained using 10- fold cross validation in order to derive a classifier. Preliminary results (over 98 % of classification rate) are satisfactory and will justify a deep investigation on how to apply mEMD for face recognition.

Escuela, familia de origen inmigrante y participación

Las relaciones entre las familias y la escuela se inscriben en la articulación entre dos instituciones con asimetría de poder y en un contexto social y político que las sitúa en el debate entre intereses públicos y privados. Aunque deben considerarse espacios yuxtapuestos, a menudo lo que se percibe es la separación, la distancia, cuando no el conflicto, entre ambos. Y esto comporta que el territorio de la escuela y el de la familia se vigile, se controle, por la amenaza de invasión o intrusión. El artículo analiza la participación de los progenitores de origen inmigrante en la escuela en España.Realizando una breve referencia a la legislación, se centra en la situación organizativa confederal, federal y asociativa (utilizando como fuente de información datos propios obtenidos en cinco grupos de discusión organizados en los diferentes niveles organizativos) y, por último, se aproxima la realidad de las Asociaciones de Padres de Alumnos (a través de una encuesta a 594 presidentes de asociaciones).Además de constatar la baja participación general y, en particular, la de las familias de origen inmigrante (menor entre unos orígenes que entre otros) se evidencia la necesidad de trabajar para incorporarlos al movimiento de padres, hecho que se considera imprescindible para su desarrollo.

APOE status modulates the changes in network connectivity induced by brain stimulation in non-demented elders.

Behavioral consequences of a brain insult represent an interaction between the injury and the capacity of the rest of the brain to adapt to it. We provide experimental support for the notion that genetic factors play a critical role in such adaptation. We induced a controlled brain disruption using repetitive transcranial magnetic stimulation (rTMS) and show that APOE status determines its impact on distributed brain networks as assessed by functional MRI (fMRI).Twenty non-demented elders exhibiting mild memory dysfunction underwent two fMRI studies during face-name encoding tasks (before and after rTMS). Baseline task performance was associated with activation of a network of brain regions in prefrontal, parietal, medial temporal and visual associative areas. APOE ε4 bearers exhibited this pattern in two separate independent components, whereas ε4-non carriers presented a single partially overlapping network. Following rTMS all subjects showed slight ameliorations in memory performance, regardless of APOE status. However, after rTMS APOE ε4-carriers showed significant changes in brain network activation, expressing strikingly similar spatial configuration as the one observed in the non-carrier group prior to stimulation. Similarly, activity in areas of the default-mode network (DMN) was found in a single component among the ε4-non bearers, whereas among carriers it appeared disaggregated in three distinct spatiotemporal components that changed to an integrated single component after rTMS. Our findings demonstrate that genetic background play a fundamental role in the brain responses to focal insults, conditioning expression of distinct brain networks to sustain similar cognitive performance.