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

Cohesion, competition and contradiction: INTERREG and Franco-British Cross border co-operation

Aquest article resumeix el desenvolupament del programa INTERREG i la cooperació a la frontera franc-britànica en els últims vint anys. Això és seguit per una anàlisi de la forma transfronterera cooperació pot entendre recorrent a les discussions conceptuals de canviar la governança europea i les relacions de poder entre els diferents nivells de govern. Aquestes idees conceptuals proporcionen un context per a un examen en profunditat de la iniciativa INTERREG IIIA, a la frontera franc-britànica que es destaquen alguns dels principals problemes i les contradiccions de la cooperació transfronterera.

Effect of Dexamethasone on cytolkine secretion in CD4+T cells in steroid refractori uveitis

Projecte de recerca elaborat a partir d’una estada a la facultat de Medical Sciences de la Universitat de Bristol, Alemanya, entre 2011 i 2012. Aquest treball s'ha realitzat a la facultat de Medical Sciences de la Universitat de Bristol, al laboratori del Prof. Andrew Di cks,o ta la seva supe1visi6. Ei treball s'ha realitzat amb els seus col.laboradors, el Dr Richard Lee i Lauren Schew:tz. Objectiu: Els glucocorticoids (GCs) tenen diversos efectes sobre les cèl.lules T CD4+ per modular la resposta immune principalment mitjançant els seus efectes anti-proliferatius. Tot i això la dexametasona (Dex, glucocorticoid sintètic) també indueix la secreció de la citocina immunosupressora IL-10 . L'objectiu d'aquest treball ha estat comparar la capacitat dels glucocorticoids en modular la producció de citocines en cèl.lules T CD4+ en pacients uveítics sensibles (SS), i resistents (SR) a esteroids. Metodologia: Es van aïllar cèl• lules T CD4+ de pacients uveítics SS i SR. Es va induir la producció de cèl.lules T regulatories (Tregs) i mitjançant I'estimulació amb anti- CD3/CD28 en presència d'lL-2 i Després del cultiu es van analitzar els nivells d’expressió intracel•lular de les citocines IL-10, IL-4, IL-9, IL-17 i IFN-y per citometria de flux. D'altra banda, també es van separar cèl.lules T CD4t de pacients uveïtis segons I'expressió de CCR6 i es van polaritzar per obtenir els fenotips ThO i Th17 per estudiar I'efecte de Dex i ciclosporina (CsA) en aquests subtipus cel.lulars. Resultats: Les cèl.lules T CD4+ de pacients SR no van ser capaces de produir IL-10 en resposta al tractament amb Dex. Dex no va afectar els nivells d'expressió d'11-17, però va reduir els nivells de IL-4 i IFN-V. Els nivells d'lL-9 (marcador d'un subtipus cel.lular recentment descrit, Th9) v ise r sempre inferiors a 11%. En canvi, el traclament a amb CsA va reduir significativament els nivells d'lL-17 i IFN-y en cèl.lules Th17 i ThO. Conclusions: La Dex no és capaç d'induir cèl.lules Treg funcionalment supresores en pacients veiticsS R. Aquest fenòmen és Independent dels efectes en I'expressió d'altres citocines. Aquests resultats suggereixen que I'efecte de la Dex sobre la funció de cél.lules Treg és clau en el desenvolupament del fenotip SR en la uveïtis . D'altra banda, al llarg d'aquest temps he iniciat un nou projecte que ha donat lloc a un futur projecte de col elaboració. Resumidament, degut a que els nivells elevats de proteïna C-reactiva (CRP) són un factor de risc en la degeneració macular, malaltia inflamatòria crònica principal causa de ceguera en països industrialitzats, I'objectiu d'aquest altre treball ha estat iniciar un projecte per avaluar els efectes de les diferents isoformes de la CRP sobre la resposta inflamatòria d’epiteli pigmentari retinià.

Exploring unfrequent events with accelerated molecular dynamics simulations: from photochemistry to drug design

La meva incorporació al grup de recerca del Prof. McCammon (University of California San Diego) en qualitat d’investigador post doctoral amb una beca Beatriu de Pinós, va tenir lloc el passat 1 de desembre de 2010; on vaig dur a terme les meves tasques de recerca fins al darrer 1 d’abril de 2012. El Prof. McCammon és un referent mundial en l’aplicació de simulacions de dinàmica molecular (MD) en sistemes biològics d’interès humà. La contribució més important del Prof. McCammon en la simulació de sistemes biològics és el desenvolupament del mètode de dinàmiques moleculars accelerades (AMD). Les simulacions MD convencionals, les quals estan limitades a l’escala de temps del nanosegon (~10-9s), no son adients per l’estudi de sistemes biològics rellevants a escales de temps mes llargues (μs, ms...). AMD permet explorar fenòmens moleculars poc freqüents però que son clau per l’enteniment de molts sistemes biològics; fenòmens que no podrien ser observats d’un altre manera. Durant la meva estada a la “University of California San Diego”, vaig treballar en diferent aplicacions de les simulacions AMD, incloent fotoquímica i disseny de fàrmacs per ordinador. Concretament, primer vaig desenvolupar amb èxit una combinació dels mètodes AMD i simulacions Car-Parrinello per millorar l’exploració de camins de desactivació (interseccions còniques) en reaccions químiques fotoactivades. En segon lloc, vaig aplicar tècniques estadístiques (Replica Exchange) amb AMD en la descripció d’interaccions proteïna-lligand. Finalment, vaig dur a terme un estudi de disseny de fàrmacs per ordinador en la proteïna-G Rho (involucrada en el desenvolupament de càncer humà) combinant anàlisis estructurals i simulacions AMD. Els projectes en els quals he participat han estat publicats (o estan encara en procés de revisió) en diferents revistes científiques, i han estat presentats en diferents congressos internacionals. La memòria inclosa a continuació conté més detalls de cada projecte esmentat.

Debt and deficit fluctuations and the structure of bond markets

This paper tests for the market environment within which US fiscal policyoperates, that is we test for the incompleteness of the US government bondmarket. We document the stochastic properties of US debt and deficits andthen consider the ability of competing optimal tax models to account forthis behaviour. We show that when a government pursues an optimal taxpolicy and issues a full set of contingent claims, the value of debthas the same or less persistence than other variables in the economyand declines in response to higher deficit shocks. By contrast, ifgovernments only issue one-period risk free bonds (incomplete markets),debt shows more persistence than other variables and it increases inresponse to expenditure shocks. Maintaining the hypothesis of Ramseybehavior, US data conflicts.

Insuring California earthquakes and the role for catastrophe bonds

The 1994 Northridge earthquake sent ripples to insurance conpanieseverywhere. This was one in a series of natural disasters such asHurricane Andrew which together with the problems in Lloyd's of Londonhave insurance companies running for cover. This paper presents a calibration of the U.S. economy in a model with financial markets forinsurance derivatives that suggests the U.S. economy can deal with thedamage of natural catastrophe far better than one might think.

Campaign rhetoric: A model of reputation

We analyze conditions under which a candidate's campaignrhetoric may affect the beliefs of the voters over whatpolicy the candidate will implement in case he wins theelection. We develop a model of repeated elections withcomplete information in which candidates are purely ideological. Voter's strategies involve a credible threat to punish candidates that renege of their campaignpromises, and in equilibrium all campaign promises arebelieved by voters, and honored by candidates. We obtainthat the degree to which promises are credible in equilibriumis an increasing function of the value of a candidate'sreputation.

Allowing for heterogeneity in the decomposition of measures of inequality in health

This paper shows how recently developed regression-based methods for the decomposition ofhealth inequality can be extended to incorporate heterogeneity in the responses of health to the explanatory variables. We illustrate our method with an application to the GHQ measure of psychological well-being taken from the British Household Panel Survey. The results suggest that there is an important degree of heterogeneity in the association of health to explanatory variables across birth cohorts and genders which, in turn, accounts for a substantial percentage of the inequality in observed health.

Measurement and explanation of socioeconomic inequality in health with longitudinal data

This paper presents a method for the measurement of changes in health inequality and income-related health inequality over time in a population.For pure health inequality (as measured by the Gini coefficient) andincome-related health inequality (as measured by the concentration index),we show how measures derived from longitudinal data can be related tocross section Gini and concentration indices that have been typicallyreported in the literature to date, along with measures of health mobilityinspired by the literature on income mobility. We also show how thesemeasures of mobility can be usefully decomposed into the contributions ofdifferent covariates. We apply these methods to investigate the degree ofincome-related mobility in the GHQ measure of psychological well-being inthe first nine waves of the British Household Panel Survey (BHPS). Thisreveals that dynamics increase the absolute value of the concentrationindex of GHQ on income by 10%.

Ambiguity in election games

We construct a model in which the ambiguity of candidates allows them toincrease the number of voters to whom they appeal when voters have intense preferences for one of the alternatives available. An ambiguous candidate may offer voters with different preferences the hope that their most preferred alternative will be implemented. We find conditions under which ambiguous strategies are chosen in equilibrium. These conditions include the case in which there is an outcome that is a majority winner against all other outcomes but is not the most preferred outcome for a majority of voters. It is shown that if the number of candidates or parties increases, ambiguity will not be possible in equilibrium, but a larger set of possible policies increases the chance that at least one candidate will choose to be ambiguous in equilibrium.

New evidence on inflation persistence and price stickiness in the Euro area: Implications for macro modelling

This paper evaluates new evidence on price setting practices and inflation persistence in the euro area with respect to its implications for macro modelling. It argues that several of the most commonly used assumptions in micro-founded macro models are seriously challenged by the new findings.

Adaptive model selection using empirical complexities

Given $n$ independent replicates of a jointly distributed pair $(X,Y)\in {\cal R}^d \times {\cal R}$, we wish to select from a fixed sequence of model classes ${\cal F}_1, {\cal F}_2, \ldots$ a deterministic prediction rule $f: {\cal R}^d \to {\cal R}$ whose risk is small. We investigate the possibility of empirically assessingthe {\em complexity} of each model class, that is, the actual difficulty of the estimation problem within each class. The estimated complexities are in turn used to define an adaptive model selection procedure, which is based on complexity penalized empirical risk.The available data are divided into two parts. The first is used to form an empirical cover of each model class, and the second is used to select a candidate rule from each cover based on empirical risk. The covering radii are determined empirically to optimize a tight upper bound on the estimation error. An estimate is chosen from the list of candidates in order to minimize the sum of class complexity and empirical risk. A distinguishing feature of the approach is that the complexity of each model class is assessed empirically, based on the size of its empirical cover.Finite sample performance bounds are established for the estimates, and these bounds are applied to several non-parametric estimation problems. The estimates are shown to achieve a favorable tradeoff between approximation and estimation error, and to perform as well as if the distribution-dependent complexities of the model classes were known beforehand. In addition, it is shown that the estimate can be consistent,and even possess near optimal rates of convergence, when each model class has an infinite VC or pseudo dimension.For regression estimation with squared loss we modify our estimate to achieve a faster rate of convergence.

The importance of individual heterogeneity in the decomposition of measures of socioeconomic inequality in health: An approach based on quantile regression

This paper shows how recently developed regression-based methods for thedecomposition of health inequality can be extended to incorporateindividual heterogeneity in the responses of health to the explanatoryvariables. We illustrate our method with an application to the CanadianNPHS of 1994. Our strategy for the estimation of heterogeneous responsesis based on the quantile regression model. The results suggest that thereis an important degree of heterogeneity in the association of health toexplanatory variables which, in turn, accounts for a substantial percentageof inequality in observed health. A particularly interesting finding isthat the marginal response of health to income is zero for healthyindividuals but positive and significant for unhealthy individuals. Theheterogeneity in the income response reduces both overall health inequalityand income related health inequality.

Homotopy Batalin-Vilkovisky algebras

This paper provides an explicit cofibrant resolution of the operad encoding Batalin-Vilkovisky algebras. Thus it defines the notion of homotopy Batalin-Vilkovisky algebras with the required homotopy properties. To define this resolution we extend the theory of Koszul duality to operads and properads that are defind by quadratic and linear relations. The operad encoding Batalin-Vilkovisky algebras is shown to be Koszul in this sense. This allows us to prove a Poincare-Birkhoff-Witt Theorem for such an operad and to give an explicit small quasi-free resolution for it. This particular resolution enables us to describe the deformation theory and homotopy theory of BV-algebras and of homotopy BV-algebras. We show that any topological conformal field theory carries a homotopy BV-algebra structure which lifts the BV-algebra structure on homology. The same result is proved for the singular chain complex of the double loop space of a topological space endowed with an action of the circle. We also prove the cyclic Deligne conjecture with this cofibrant resolution of the operad BV. We develop the general obstruction theory for algebras over the Koszul resolution of a properad and apply it to extend a conjecture of Lian-Zuckerman, showing that certain vertex algebras have an explicit homotopy BV-algebra structure.