Pauvreté | EPFL Graph Search

alt=|vignette|250x250px|Manifestation contre la pauvreté des étudiants, France. alt=|vignette|250x250px|Habitants des rues du district de San'ya à Tokyo, au Japon. La pauvreté désigne dans une société donnée le fait d'être dans une situation d'infériorité matérielle par rapport aux individus les plus favorisés; cela se traduit notamment par des difficultés à subvenir à ses besoins et à ceux de ses proches, mais aussi par une stigmatisation de la part des personnes plus riches. Il n'existe cependant pas de définition complètement consensuelle et universelle de la pauvreté. La satisfaction des besoins de base est jugée comme indispensable à la vie décente d'un être humain; avec le progrès technique et l'amélioration des conditions de vie dans les pays développés, une définition basée sur des seuils de pauvreté relatifs au revenu médian a vu le jour sans lien avec la satisfaction de ces besoins. Autrement dit, le nombre de pauvr

On Vacuous and Non-Vacuous Generalization Bounds for Deep Neural Networks.

Konstantinos Pitas

Deep neural networks have been empirically successful in a variety of tasks, however their theoretical understanding is still poor. In particular, modern deep neural networks have many more parameters than training data. Thus, in principle they should overfit the training samples and exhibit poor generalization to the complete data distribution. Counter intuitively however, they manage to achieve both high training accuracy and high testing accuracy. One can prove generalization using a validation set, however this can be difficult when training samples are limited and at the same time we do not obtain any information about why deep neural networks generalize well. Another approach is to estimate the complexity of the deep neural network. The hypothesis is that if a network with high training accuracy has high complexity it will have memorized the data, while if it has low complexity it will have learned generalizable patterns. In the first part of this thesis we explore Spectral Complexity, a measure of complexity that depends on combinations of norms of the weight matrices of the deep neural network. For a dataset that is difficult to classify, with no underlying model and/or no recurring pattern, for example one where the labels have been chosen randomly, spectral complexity has a large value, reflecting that the network needs to memorize the labels, and will not generalize well. Putting back the real labels, the spectral complexity becomes lower reflecting that some structure is present and the network has learned patterns that might generalize to unseen data. Spectral complexity results in vacuous estimates of the generalization error (the difference between the training and testing accuracy), and we show that it can lead to counterintuitive results when comparing the generalization error of different architectures. In the second part of the thesis we explore non-vacuous estimates of the generalization error. In Chapter 2 we analyze the case of PAC-Bayes where a posterior distribution over the weights of a deep neural network is learned using stochastic variational inference, and the generalization error is the KL divergence between this posterior and a prior distribution. We find that a common approximation where the posterior is constrained to be Gaussian with diagonal covariance, known as the mean-field approximation, limits significantly any gains in bound tightness. We find that, if we choose the prior mean to be the random network initialization, the generalization error estimate tightens significantly. In Chapter 3 we explore an existing approach to learning the prior mean, in PAC-Bayes, from the training set. Specifically, we explore differential privacy, which ensures that the training samples contribute only a limited amount of information to the prior, making it distribution and not training set dependent. In this way the prior should generalize well to unseen data (as it hasn't memorized individual samples) and at the same time any posterior distribution that is close to it in terms of the KL divergence will also exhibit good generalization.

EPFL2020

Sustainable Development, Planning and Poverty Alleviation

Jean-Claude Bolay, Eléonore Labattut

In 2018, the world population is around 7.6 billion, 4.2 billion in urban settlements and 3.4 billion in rural areas. Of this total, according to UN-Habitat, 3.2 billion of urban inhabitants live in southern countries. Of them, one billion, or nearly a third, live in slums. Urban poverty is therefore an endemic problem that has not been solved despite all initiatives taken to date by public and private sectors. This global transformation of our contemporary societies is particularly challenging in Asia and Africa, knowing that on these two continents, less than half of the population currently lives in urban areas. In addition, over the next decades, 90% of the urbanization process will take place in these major regions of the world. Urban planning is not an end in itself. It is a way, human and technological, to foresee the future and to act in a consistent and responsible way in order to guarantee the well-being of the populations residing in cities or in their peripheries. Many writers and urban actors in the South have criticized the inadequacy of urban planning to the problems faced by the cities confronting spatial and demographic growth. For many of them the reproduction of Western models of planning is ineffective when the urban context responds to very different logics. It is therefore a question of reinventing urban planning on different bases. And in order to address the real problems that urban inhabitants and authorities are facing, and offering infrastructures and access to services for all, this with the prospect of reducing poverty, to develop a more inclusive city, with a more efficient organization, in order to make it sustainable, both environmental than social and economic. The field work carried out during recent years in small and medium-sized cities in Burkina Faso, Brazil, Argentina and Vietnam allows us to focus the attention of specialists and decision makers on intermediate cities that have been little studied but which are home to half of the world's urban population. From local diagnoses, we come to a first conclusion. Many small and medium-sized cities in the South can be considered as poor cities, from four criteria. They have a relatively large percentage of the population is considered to be poor; the local government and its administration do not have enough money to invest in solving the problems they face; these same authorities lack the human resources to initiate and manage an efficient planning process; urban governance remains little open to democratic participation and poorly integrates social demand into its development plans. Based on this analysis, we consider it is imperative to renovate urban planning as part of a more participatory process that meets the expectations of citizens with more realistic criteria. This process incorporates different stages: an analysis grounded on the identification of urban investment needed to improve the city; the consideration of the social demands; a realistic assessment of the financial resources to be mobilized (municipal budget, taxes, public and international external grants, public private partnership); a continuous dialogue between urban actors to determine the urban priorities to be addressed in the coming years. This protocol serves as a basis for comparative studies between cities in the South and a training program initiated in Argentina for urban actors in small and medium-sized cities, which we wish to extend later to other countries of the South.

2019

Visualization of hepatic arteries with 3D ultrasound during intra-arterial therapies

Anaïs Laure Marie-Thérèse Badoual, Alexandre Bigot, François Michaud

Liver cancer represents the second most common cause of cancer-related mortality worldwide. The prognosis is poor with an overall mortality of 95%. Moreover, most hepatic tumors are unresectable due to their advanced stage at discovery or poor underlying liver function. Tumor embolization by intra-arterial approaches is the current standard of care for advanced cases of hepatocellular carcinoma. These therapies rely on the fact that the blood supply of primary hepatic tumors is predominantly arterial. Feedback on blood flow velocities in the hepatic arteries is crucial to ensure maximal treatment efficacy on the targeted masses. Based on these velocities, the intra-arterial injection rate is modulated for optimal infusion of the chemotherapeutic drugs into the tumorous tissue. While Doppler ultrasound is a well-documented technique for the assessment of blood flow, 3D visualization of vascular anatomy with ultrasound remains challenging. In this paper we present an image-guidance pipeline that enables the localization of the hepatic arterial branches within a 3D ultrasound image of the liver. A diagnostic Magnetic resonance angiography (MRA) is first processed to automatically segment the hepatic arteries. A non-rigid registration method is then applied on the portal phase of the MRA volume with a 3D ultrasound to enable the visualization of the 3D mesh of the hepatic arteries in the Doppler images. To evaluate the performance of the proposed workflow, we present initial results from porcine models and patient images.