Datasets, features, learning, and models in visual recognition

Visual recognition is a fundamental research topic in computer vision. This dissertation explores datasets, features, learning, and models used for visual recognition. In order to train visual models and evaluate different recognition algorithms, this dissertation develops an approach to collect object image datasets on web pages using an analysis of text around the image and of image appearance. This method exploits established online knowledge resources (Wikipedia pages for text; Flickr and Caltech data sets for images). The resources provide rich text and object appearance information. This dissertation describes results on two datasets. The first is Berg’s collection of 10 animal categories; on this dataset, we significantly outperform previous approaches. On an additional set of 5 categories, experimental results show the effectiveness of the method. Images are represented as features for visual recognition. This dissertation introduces a text-based image feature and demonstrates that it consistently improves performance on hard object classification problems. The feature is built using an auxiliary dataset of images annotated with tags, downloaded from the Internet. Image tags are noisy. The method obtains the text features of an unannotated image from the tags of its k-nearest neighbors in this auxiliary collection. A visual classifier presented with an object viewed under novel circumstances (say, a new viewing direction) must rely on its visual examples. This text feature may not change, because the auxiliary dataset likely contains a similar picture. While the tags associated with images are noisy, they are more stable when appearance changes. The performance of this feature is tested using PASCAL VOC 2006 and 2007 datasets. This feature performs well; it consistently improves the performance of visual object classifiers, and is particularly effective when the training dataset is small. With more and more collected training data, computational cost becomes a bottleneck, especially when training sophisticated classifiers such as kernelized SVM. This dissertation proposes a fast training algorithm called Stochastic Intersection Kernel Machine (SIKMA). This proposed training method will be useful for many vision problems, as it can produce a kernel classifier that is more accurate than a linear classifier, and can be trained on tens of thousands of examples in two minutes. It processes training examples one by one in a sequence, so memory cost is no longer the bottleneck to process large scale datasets. This dissertation applies this approach to train classifiers of Flickr groups with many group training examples. The resulting Flickr group prediction scores can be used to measure image similarity between two images. Experimental results on the Corel dataset and a PASCAL VOC dataset show the learned Flickr features perform better on image matching, retrieval, and classification than conventional visual features. Visual models are usually trained to best separate positive and negative training examples. However, when recognizing a large number of object categories, there may not be enough training examples for most objects, due to the intrinsic long-tailed distribution of objects in the real world. This dissertation proposes an approach to use comparative object similarity. The key insight is that, given a set of object categories which are similar and a set of categories which are dissimilar, a good object model should respond more strongly to examples from similar categories than to examples from dissimilar categories. This dissertation develops a regularized kernel machine algorithm to use this category dependent similarity regularization. Experiments on hundreds of categories show that our method can make significant improvement for categories with few or even no positive examples.

Econometric models in foreign exchange market

According to the significance of the econometric models in foreign exchange market, the purpose of this research is to give a closer examination on some important issues in this area. The research covers exchange rate pass-through into import prices, liquidity risk and expected returns in the currency market, and the common risk factors in currency markets. Firstly, with the significant of the exchange rate pass-through in financial economics, the first empirical chapter studies on the degree of exchange rate pass-through into import in emerging economies and developed countries in panel evidences for comparison covering the time period of 1970-2009. The pooled mean group estimation (PMGE) is used for the estimation to investigate the short run coefficients and error variance. In general, the results present that the import prices are affected positively, though incompletely, by the exchange rate. Secondly, the following study addresses the question whether there is a relationship between cross-sectional differences in foreign exchange returns and the sensitivities of the returns to fluctuations in liquidity, known as liquidity beta, by using a unique dataset of weekly order flow. Finally, the last study is in keeping with the study of Lustig, Roussanov and Verdelhan (2011), which shows that the large co-movement among exchange rates of different currencies can explain a risk-based view of exchange rate determination. The exploration on identifying a slope factor in exchange rate changes is brought up. The study initially constructs monthly portfolios of currencies, which are sorted on the basis of their forward discounts. The lowest interest rate currencies are contained in the first portfolio and the highest interest rate currencies are in the last. The results performs that portfolios with higher forward discounts incline to contain higher real interest rates in overall by considering the first portfolio and the last portfolio though the fluctuation occurs.

Problems in number theory and hyperbolic geometry

In the first part of this thesis we generalize a theorem of Kiming and Olsson concerning the existence of Ramanujan-type congruences for a class of eta quotients. Specifically, we consider a class of generating functions analogous to the generating function of the partition function and establish a bound on the primes ℓ for which their coefficients c(n) obey congruences of the form c(ℓn + a) ≡ 0 (mod ℓ). We use this last result to answer a question of H.C. Chan. In the second part of this thesis [S2] we explore a natural analog of D. Calegari’s result that there are no hyperbolic once-punctured torus bundles over S^1 with trace field having a real place. We prove a contrasting theorem showing the existence of several infinite families of pairs (−χ, p) such that there exist hyperbolic surface bundles over S^1 with trace field of having a real place and with fiber having p punctures and Euler characteristic χ. This supports our conjecture that with finitely many known exceptions there exist such examples for each pair ( −χ, p).

Local Field Potentials in the Gustatory Cortex Carry Taste Information

It has been recently shownthat localfield potentials (LFPs)fromthe auditory and visual cortices carry information about sensory stimuli, but whether this is a universal property of sensory cortices remains to be determined. Moreover, little is known about the temporal dynamics of sensory information contained in LFPs following stimulus onset. Here we investigated the time course of the amount of stimulus information in LFPs and spikes from the gustatory cortex of awake rats subjected to tastants and water delivery on the tongue. We found that the phase and amplitude of multiple LFP frequencies carry information about stimuli, which have specific time courses after stimulus delivery. The information carried by LFP phase and amplitude was independent within frequency bands, since the joint information exhibited neither synergy nor redundancy. Tastant information in LFPs was also independent and had a different time course from the information carried by spikes. These findings support the hypothesis that the brain uses different frequency channels to dynamically code for multiple features of a stimulus.

The use of animal models in multiple myeloma

International audience

A new method for conditioning stochastic groundwater flow models in fractured media

Many geological formations consist of crystalline rocks that have very low matrix permeability but allow flow through an interconnected network of fractures. Understanding the flow of groundwater through such rocks is important in considering disposal of radioactive waste in underground repositories. A specific area of interest is the conditioning of fracture transmissivities on measured values of pressure in these formations. This is the process where the values of fracture transmissivities in a model are adjusted to obtain a good fit of the calculated pressures to measured pressure values. While there are existing methods to condition transmissivity fields on transmissivity, pressure and flow measurements for a continuous porous medium there is little literature on conditioning fracture networks. Conditioning fracture transmissivities on pressure or flow values is a complex problem because the measurements are not linearly related to the fracture transmissivities and they are also dependent on all the fracture transmissivities in the network. We present a new method for conditioning fracture transmissivities on measured pressure values based on the calculation of certain basis vectors; each basis vector represents the change to the log transmissivity of the fractures in the network that results in a unit increase in the pressure at one measurement point whilst keeping the pressure at the remaining measurement points constant. The fracture transmissivities are updated by adding a linear combination of basis vectors and coefficients, where the coefficients are obtained by minimizing an error function. A mathematical summary of the method is given. This algorithm is implemented in the existing finite element code ConnectFlow developed and marketed by Serco Technical Services, which models groundwater flow in a fracture network. Results of the conditioning are shown for a number of simple test problems as well as for a realistic large scale test case.

Changing family models in Spain: the impact of the sociodemographic factors

Abstract During the last few decades, there has been an increasing international recognition of the studies related to the analysis of the family models change, the focus being the determinants of the female employment and the problems related to the work family balance (Lewis, 2001; Petit & Hook, 2005Saraceno, Crompton & Lyonette, 20062008; Pfau-Effinger, 2012). The majority of these studies have been focused on the analysis of the work-family balance problems as well as the effectiveness of the family and gender policies in order to encourage female employment (Korpi et al., 2013). In Spain, special attention has been given to the family policies implemented, the employability of women and on the role of the father in the family (Flaquer et al., 2015; Meil, 2015); however, there has been far less emphasis on the analysis of the family cultural models (González and Jurado, 2012; Crespi and Moreno, 2016). The purpose of this paper is to present some of the first results on the influence of the socio-demographic factors on the expectations and attitudes about the family models. This study offers an analytical reflection upon the foundation of the determinants of the family ambivalence in Spain from the cultural and the institutional dimension. This study shows the Spanish family models of preferences following the Pfau-Effinger (2004) classification of the famiy living arrangements. The reason for this study is twofold; on the one hand, there is confirmed the scarcity of studies that have focused their attention on this objective in Spain; on the other hand, the studies carried out in the international context have confirmed the analytical effectiveness of researching on the attitude and value changes to explain the meaning and trends of the family changes. There is also presented some preliminary results that have been obtained from the multinomial analysis related to the influence of the socio-demographic factors on the family model chosen by the individuals in Spain (father and mother working full time; mother part-time father full-time; mother not at work father full-time; mother and father part-time). 3 The database used has been the International Social Survey Programme: Family and Changing Gender Roles IV- ISSP 2012-. Spain is the only country of South Europe that has participated in the survey. For this reason it has been considered as a representative case study.

Transferability of environmental favourability models in geographic space: The case of the Iberian desman (Galemys pyrenaicus) in Portugal and Spain

Transferring distribution models between different geographical areas may be problematic, as the performance of models outside their original scope is hard to predict. A modelling procedure is needed that gets the gist of the environmental descriptors of a distribution area, without either overfitting to the training data or overestimating the species’ distribution potential.We tested the transferability power of the favourability function, a generalized linear model, on the distribution of the Iberian desman (Galemys pyrenaicus) in the Iberian territories of Portugal and Spain.We also tested the effects of two of the main potential constraints on model transferability: the analysed ranges of the predictor variables, and the completeness of the species distribution data. We modelled 10 km×10km presence/absence data from Portugal and Spain separately, extrapolated each model to the other country, and compared predictions with observations. The Spanish model, despite arguably containing more false absences, showed good predictive ability in Portugal. The Portuguese model, whose predictors ranged between only a subset of the values observed in Spain, overestimated desman distribution when transferred.We discuss possible reasons for this differential model behaviour, and highlight the importance of this kind of models for prediction and conservation applications

Spatial scale effects of sampling on the interpolation of species distribution models in the southwestern Amazon.

Knowledge of the geographical distribution of timber tree species in the Amazon is still scarce. This is especially true at the local level, thereby limiting natural resource management actions. Forest inventories are key sources of information on the occurrence of such species. However, areas with approved forest management plans are mostly located near access roads and the main industrial centers. The present study aimed to assess the spatial scale effects of forest inventories used as sources of occurrence data in the interpolation of potential species distribution models. The occurrence data of a group of six forest tree species were divided into four geographical areas during the modeling process. Several sampling schemes were then tested applying the maximum entropy algorithm, using the following predictor variables: elevation, slope, exposure, normalized difference vegetation index (NDVI) and height above the nearest drainage (HAND). The results revealed that using occurrence data from only one geographical area with unique environmental characteristics increased both model overfitting to input data and omission error rates. The use of a diagonal systematic sampling scheme and lower threshold values led to improved model performance. Forest inventories may be used to predict areas with a high probability of species occurrence, provided they are located in forest management plan regions representative of the environmental range of the model projection area.

Entanglement entropy from the holographic stress tensor

We consider entanglement entropy in the context of gauge/gravity duality for conformal field theories in even dimensions. The holographic prescription due to Ryu and Takayanagi (RT) leads to an equation describing how the entangling surface extends into the bulk geometry. We show that setting to zero, the timetime component of the Brown-York stress tensor evaluated on the co-dimension 1 entangling surface, leads to the same equation. By considering a spherical entangling surface as an example, we observe that the Euclidean actionmethods in AdS/CFT will lead to the RT area functional arising as a counterterm needed to regularize the stress tensor. We present arguments leading to a justification for the minimal area prescription.

Desingularization of the Milne Universe

Resolution of cosmological singularities is an important problem in any full theory of quantum gravity. The Milne orbifold is a cosmology with a big-bang/big-crunch singularity, but being a quotient of flat space it holds potential for resolution in string theory. It is known, however, that some perturbative string amplitudes diverge in the Milne geometry. Here we show that flat space higher spin theories can effect a simple resolution of the Milne singularity when one embeds the latter in 2 + 1 dimensions. We explain how to reconcile this with the expectation that non-perturbative string effects are required for resolving Milne. Along the way, we introduce a Grassmann realization of the inonfi-Wigner contraction to export much of the AdS technology to -our flat space computation. (C) 2014 The Authors. Published by Elsevier BAT.

Low tension strings on a cosmological singularity

It has recently been argued that the singularity of the Milne orbifold can be resolved in higher spin theories. In string theory scattering amplitudes, however, the Milne singularity gives rise to ultraviolet divergences that signal uncontrolled backreaction. Since string theory in the low tension limit is expected to be a higher spin theory (although precise proposals only exist in special cases), we investigate what happens to these scattering amplitudes in the low tension limit. We point out that the known problematic ultraviolet divergences disappear in this limit. In addition we systematically identify all divergences of the simplest 2-to-2 tree-level string scattering amplitude on the Milne orbifold, and argue that the divergences that survive in the low tension limit have sensible infrared interpretations.

Analytic bootstrap at large spin

We use analytic conformal bootstrap methods to determine the anomalous dimensions and OPE coefficients for large spin operators in general conformal field theories in four dimensions containing a scalar operator of conformal dimension Delta(phi). It is known that such theories will contain an in finite sequence of large spin operators with twists approaching 2 Delta(phi) + 2n for each integer n. By considering the case where such operators are separated by a twist gap from other operators at large spin, we analytically determine the n, Delta(phi) dependence of the anomalous dimensions. We find that for all n, the anomalous dimensions are negative for Delta(phi) satisfying the unitarity bound. We further compute the first subleading correction at large spin and show that it becomes universal for large twist. In the limit when n is large, we find exact agreement with the AdS/CFT prediction corresponding to the Eikonal limit of a 2-2 scattering with dominant graviton exchange.

Characterization of Probability Distributions Using the Residual Entropy Function

The present study on the characterization of probability distributions using the residual entropy function. The concept of entropy is extensively used in literature as a quantitative measure of uncertainty associated with a random phenomenon. The commonly used life time models in reliability Theory are exponential distribution, Pareto distribution, Beta distribution, Weibull distribution and gamma distribution. Several characterization theorems are obtained for the above models using reliability concepts such as failure rate, mean residual life function, vitality function, variance residual life function etc. Most of the works on characterization of distributions in the reliability context centers around the failure rate or the residual life function. The important aspect of interest in the study of entropy is that of locating distributions for which the shannon’s entropy is maximum subject to certain restrictions on the underlying random variable. The geometric vitality function and examine its properties. It is established that the geometric vitality function determines the distribution uniquely. The problem of averaging the residual entropy function is examined, and also the truncated form version of entropies of higher order are defined. In this study it is established that the residual entropy function determines the distribution uniquely and that the constancy of the same is characteristics to the geometric distribution

Black hole entropy associated with the supersymmetric sigma model

By means of an identity that equates the elliptic genus partition function of a supersymmetric sigma model on the N-fold symmetric product (SX)-X-N of X ((SX)-X-N=X-N/S-N, where S-N is the symmetric group of N elements) to the partition function of a second-quantized string theory, we derive the asymptotic expansion of the partition function as well as the asymptotic for the degeneracy of spectrum in string theory. The asymptotic expansion for the state counting reproduces the logarithmic correction to the black hole entropy.