Letteratura scientifica selezionata sul tema "Module learning with errors"

La transition vers la cryptographie post-quantique est une tâche considérable ayant suscité un nombre important de travaux ces dernières années. En parallèle, la cryptographie pour la protection de la vie privée, visant à pallier aux limitations inhérentes des mécanismes cryptographiques basiques dans ce domaine, a connu un véritable essor. Malgré le succès de chacune de ces branches prises individuellement, combiner les deux aspects de manière efficace s'avère extrêmement difficile. Le but de cette thèse de doctorat consiste alors à proposer de nouvelles constructions visant à garantir une protection efficace et post-quantique de la vie privée, et plus généralement des mécanismes d'authentification avancés. Dans ce but, nous nous consacrons tout d'abord à l'étude de l'une des hypothèses mathématiques fondamentales utilisées en cryptographie sur les réseaux Euclidiens: Module Learning With Errors. Nous prouvons que le problème ne devient pas significativement plus facile même en choisissant des distributions de secret et d'erreur plus courtes. Ensuite, nous proposons des optimisations des échantillonneurs d'antécédents utilisés par de nombreuses signatures avancées. Loin d'être limitées à ce cas d'usage, nous montrons que ces optimisations mènent à la conception de signatures standards efficaces. Enfin, à partir de ces contributions, nous concevons des algorithmes de signatures avec protocoles efficaces, un outil polyvalent utile à la construction d'applications avancées. Nous en montrons les capacités en proposant le premier mécanisme d'accréditation anonyme post-quantique, que nous implémentons afin de mettre en exergue son efficacité aussi bien théorique que pratique<br>The transition to post-quantum cryptography has been an enormous effort for cryptographers over the last decade. In the meantime, cryptography for the protection of privacy, aiming at addressing the limitations inherent to basic cryptographic mechanisms in this domain, has also attracted a lot of attention. Nevertheless, despite the success of both individual branches, combining both aspects along with practicality turns out to be very challenging. The goal of this thesis then lies in proposing new constructions for practical post-quantum privacy, and more generally advanced authentication mechanisms. To this end, we first focus on the lower level by studying one of the fundamental mathematical assumptions used in lattice-based cryptography: Module Learning With Errors. We show that it does not get significantly easier when stretching the secret and error distributions. We then turn to optimizing preimage samplers which are used in advanced signature designs. Far from being limited to this use case, we show that it also leads to efficient designs of regular signatures. Finally, we use some of the previous contributions to construct so-called signatures with efficient protocols, a versatile building block in countless advanced applications. We showcase it by giving the first post-quantum anonymous credentials, which we implement to demonstrate a theoretical and practical efficiency

Classical supervised learning from a training set of labelled examples assumes that the labels are correct. But in reality labelling errors may originate, for example, from human mistakes, diverging human opinions, or errors of the measuring instruments. In such cases the training set is misleading and in consequence the learning may suffer. In this thesis we consider probabilistic modelling of random label noise. The goal of this research is two-fold. First, to develop new improved algorithms and architectures from a principled footing which are able to detect and bypass the unwanted effects of mislabelling. Second, to study the performance of such methods both empirically and theoretically. We build upon two classical probabilistic classifiers, the normal discriminant analysis and the logistic regression and introduce the label-noise robust versions of these classifiers. We also develop useful extensions such as a sparse extension and a kernel extension in order to broaden applicability of the robust classifiers. Finally, we devise an ensemble of the robust classifiers in order to understand how the robust models perform collectively. Theoretical and empirical analysis of the proposed models show that the new robust models are superior to the traditional approaches in terms of parameter estimation and classification performance.

Thesis (M.Eng.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2001.<br>Includes bibliographical references (leaf 103).<br>Due to increased demand for interactive learning opportunities for engineering students, an interactive spectral analysis learning module was developed for the course Biomedical Signal and Image Processing (HST582J/6.555J/16.456J). The design of this module is based on the Star Legacy model, a pedagogical framework that promotes the creation of guided learning environments that use applications as the context for focused learning activities. The module is implemented using a combination of traditional teaching methods and web-based components. The web-based components include tutorial questions, text summaries, tables, figures, a glossary, and an interactive demonstration. This module was used in HST582J/6.555J/16.456J during Spring Term 2001. A variety of assessment techniques were employed. Survey results show that students generally found the module useful. Student performance on lab reports showed improved understanding of key concepts relative to previous years. Future efforts should reanalyze other performance data and make suggested modifications to the overall module, the web-based tutorial, and the interactive demo.<br>by Natalie T. Smith.<br>M.Eng.

La cryptographie à base de réseaux euclidiens repose en grande partie sur l’utilisation du problème Learning With Errors (LWE) comme fondation de sécurité. Ce problème est au moins aussi difficile que les problèmes standards portant sur les réseaux, mais les primitives cryptographiques qui l’utilisent sont inefficaces en termes de consommation en temps et en espace. Les problèmes Polynomial Learning WithErrors (PLWE), dual Ring Learning With Errors (dual-RLWE) et primal Ring Learning With Errors(primal-RLWE) sont trois variantes de LWE qui utilisent des structures algébriques supplémentaires afin de pallier les inconvénients ci-dessus. Le problème PLWE est paramétré par un polynôme f, alors que dual-RLWE et primal-RLWE sont définis à l’aide de l’anneau d’entiers d’un corps de nombres.Ces problèmes, dits algébriques, sont eux-mêmes au moins aussi difficiles que des problèmes standards portant sur les réseaux, mais, dans leur cas, les réseaux impliqués appartiennent à des classes restreintes.Dans cette thèse, nous nous intéressons aux liens entre les variantes algébriques de LWE.Tout d’abord, nous montrons que pour une vaste classe de polynômes de définition, il existe des réductions (non-uniformes) entre dual-RLWE, primal-RLWE et PLWE pour lesquelles l’amplification des paramètres peut être contrôlée. Ces résultats peuvent être interprétés comme une indication forte de l’équivalence calculatoire de ces problèmes.Ensuite, nous introduisons une nouvelle variante algébrique de LWE, Middle-Product Learning WithErrors (MP-LWE). On montre que ce problème est au moins aussi difficile que PLWE pour beaucoup de polynômes de définition f. Par conséquent, un système cryptographique reposant sur MP-LWE reste sûr aussi longtemps qu’une de ces instances de PLWE reste difficile à résoudre.Enfin, nous montrons la pertinence cryptographique de MP-LWE en proposant un protocole de chiffrement asymétrique et une signature digitale dont la sécurité repose sur la difficulté présumée de MP-LWE<br>Lattice-based cryptography relies in great parts on the use of the Learning With Errors (LWE) problemas hardness foundation. This problem is at least as hard as standard worst-case lattice problems, but the primitives based on it usually have big key sizes and slow algorithms. Polynomial Learning With Errors (PLWE), dual Ring Learning With Errors (dual-RLWE) and primal Ring Learning WithErrors (primal-RLWE) are variants of LWE which make use of extra algebraic structures in order to fix the above drawbacks. The PLWE problem is parameterized by a polynomial f, while dual-RLWE andprimal-RLWE are defined using the ring of integers of a number field. These problems, which we call algebraic, also enjoy reductions from worst-case lattice problems, but in their case, the lattices involved belong to diverse restricted classes. In this thesis, we study relationships between algebraic variants of LWE.We first show that for many defining polynomials, there exist (non-uniform) reductions betweendual-RLWE, primal-RLWE and PLWE that incur limited parameter losses. These results could be interpretedas a strong evidence that these problems are qualitatively equivalent.Then we introduce a new algebraic variant of LWE, Middle-Product Learning With Errors (MP-LWE). We show that this problem is at least as hard as PLWE for many defining polynomials f. As a consequence,any cryptographic system based on MP-LWE remains secure as long as one of these PLWE instances remains hard to solve.Finally, we illustrate the cryptographic relevance of MP-LWE by building a public-key encryption scheme and a digital signature scheme that are proved secure under the MP-LWE hardness assumption

Although errors might foster learning, they can also be perceived as something to avoid if they are associated with negative consequences (e.g., receiving a bad grade or being mocked by classmates). Such adverse perceptions may trigger negative emotions and error-avoidance attitudes, limiting the possibility to use errors for learning. These students’ reactions may be influenced by relational and cultural aspects of errors that characterise the learning environment. Accordingly, the main aim of this research was to investigate whether relational and cultural characteristics associated with errors affect psychological mechanisms triggered by making mistakes. In the theoretical part, we described the role of errors in learning using an integrated multilevel (i.e., psychological, relational, and cultural levels of analysis) approach. Then, we presented three studies that analysed how cultural and relational error-related variables affect psychological aspects. The studies adopted a specific empirical methodology (i.e., qualitative, experimental, and correlational) and investigated different samples (i.e., teachers, primary school pupils and middle school students). Findings of study one (cultural level) highlighted errors acquire different meanings that are associated with different teachers’ error-handling strategies (e.g., supporting or penalising errors). Study two (relational level) demonstrated that teachers’ supportive error-handling strategies promote students’ perceptions of being in a positive error climate. Findings of study three (relational and psychological level) showed that positive error climate foster students’ adaptive reactions towards errors and learning outcomes. Overall, our findings indicated that different variables influence students’ learning from errors process and teachers play an important role in conveying specific meanings of errors during learning activities, dealing with students’ mistakes supportively, and establishing an error-friendly classroom environment.

<p>Road traffic safety has increasingly become in need of educated road safety professionals, as the number of accidents in the World Health Organization member countries exceeds one million. The profession itself is transitioning from experience based decision making to empirical, theoretical and mathematical based solutions. However, road traffic safety is a multidiscipline, crossing over many fields and requiring a high degree of communication between different institutions. There are very few institutions that provide programs in the field; furthermore, they employ traditional lecture-based teaching methods. The traditional teaching environment does not fulfill the educational needs of future traffic safety professionals due to its rigidity and lack of problem solving exercises.</p><p>An alternative method, namely problem based learning, is recommended as an alternative teaching method in this paper. The thesis is constructed in such a way as to develop a complete road traffic safety educational module at graduate and post graduate level.</p><p>The theoretical basis on which a road traffic safety module is later built is presented in the first part of the thesis. Major concepts in road traffic safety, as well as problem based learning methods are investigated. In addition, a literature review SWOT analysis based on literature is conducted.The module development consists of establishing the road traffic safety learning goals for each segment in the module, appropriate assessment criteria and group work format. The module contains gradual difficulty level problems, starting from the easiest topic and easiest format (closed ended problem) and ending with the hardest topic and hardest format (open ended problem).</p><p>The last section employs the SWOT analysis findings in the theoretical section to develop a SWOT analysis of the road traffic safety module presented in the thesis.</p>

A informação disponível para robôs em tarefas reais encontra-se amplamente distribuída tanto no espaço quanto no tempo, fazendo com que o agente busque informações relevantes. Neste trabalho, uma solução que usa conhecimento qualitativo e quantitativo da tarefa é implementada a fim de permitir que tarefas robóticas reais sejam tratáveis por algoritmos de Aprendizagem por Reforço (AR). Os passos deste procedimento incluem: 1) decompor a tarefa completa em tarefas menores, usando abstração e macro-operadores, para que um espaço de ações discreto seja atingido; 2) aplicar um modelo de representação do espaço de estados a fim de atingir discretização tanto no espaço de estados quanto no de tempo; 3) usar conhecimento quantitativo para projetar controladores capazes de resolver as subtarefas; 4) aprender a coordenação destes comportamentos usando AR, mais especificamente o algoritmo Q-learning. O método proposto foi verificado em um conjunto de tarefas de complexidade crescente por meio de um simulador para o robô Khepera. Dois modelos de discretização para o espaço de estados foram usados, um baseado em estados e outro baseado em atributos --- funções de observação do ambiente. As políticas aprendidas sobre estes dois modelos foram comparadas a uma política pré-definida. Os resultados mostraram que a política aprendida sobre o modelo de discretização baseado em estados leva mais rapidamente a resultados melhores, apesar desta não poder ser aplicada a tarefas mais complexas, onde o espaço de estados sob esta representação se torna computacionalmente inviável e onde um método de generalização deve ser aplicado. O método de generalização escolhido implementa a estrutura CMAC ( extit{Cerebellar Model Articulation Controller}) sobre o modelo de discretização baseado em estados. Os resultados mostraram que a representação compacta permite que o algoritmo de aprendizagem seja aplicado sobre este modelo, apesar de que, para este caso, a política aprendida sob o modelo de discretização baseado em atributos apresenta melhor performance.

Dissertação para obtenção do Grau de Mestre em Engenharia Electrotécnica, Sistemas e Computadores<br>The dissertation presents the work done under the scope of the NP7 Self-Learning project regarding the design and development of the Adapter component as a foundation for the Self-Learning Production Systems (SLPS). This component is responsible to confer additional proprieties to production systems such as lifecycle learning, optimization of process parameters and, above all, adaptation to different production contexts. Therefore, the SLPS will be an evolvable system capable to self-adapt and learn in response to dynamic contextual changes in manufacturing production process in which it operates. The key assumption is that a deeper use of data mining and machine learning techniques to process the huge amount of data generated during the production activities will allow adaptation and enhancement of control and other manufacturing production activities such as energy use optimization and maintenance. In this scenario, the SLPS Adapter acts as a doer and is responsible for dynamically adapting the manufacturing production system parameters according to changing manufacturing production contexts and, most important, according to the history of the manufacturing production process acquired during SLPS run time.To do this, a Learning Module has been also developed and embedded into the SLPS Adapter. The SLPS Learning Module represents the processing unit of the SLPS Adapter and is responsible to deliver Self-learning capabilities relying on data mining and operator’s feedback to up-date the execution of adaptation and context extraction at run time. The designed and implemented SLPS Adapter architecture is assessed and validated into several application scenario provided by three industrial partners to assure industrial relevant self-learning production systems. Experimental results derived by the application of the SLPS prototype into real industrial environment are also presented.

Thesis: Ph. D. in Linguistics, Massachusetts Institute of Technology, Department of Linguistics and Philosophy, 2015.<br>Cataloged from PDF version of thesis. "September 2015."<br>Includes bibliographical references (pages 266-275).<br>In this thesis I explore the role of ambiguous evidence in first language acquisition by using a probabilistic learner for setting syntactic parameters. As ambiguous evidence is input to the learner that is compatible with multiple grammars or hypotheses, it poses learnability and acquisition challenges because it underdetermines the correct analysis. However, a probabilistic learning model with competing hypotheses can address these challenges by learning from general tendencies regarding the shape of the input, thereby finding the most compatible set of hypotheses, or the grammar with the 'best fit' to the input. This enables the model to resolve the challenge of learning the grammar of a subset language: it can reach such a target end-state by learning from implicit negative evidence. Moreover, ambiguous evidence can provide insight into two phenomena characteristic of language acquisition: variability (both within speakers and across a population) and learning errors. Both phenomena can be accounted for under a model that is attempting to learn a grammar of best fit. Three case studies relating to word order and phrase structure are investigated with simulations of the model. First I show how the model can account for embedded clause verb placement errors in child Swiss German by learning from ambiguous input. I then show how learning from ambiguous input allows the model to account for grammatical variability across speakers with regard to verb movement in Korean. Finally, I show that the model is successfully able to learn the grammar of a subset language with the example of zero-derived causatives in English.<br>by Isaac Gould.<br>Ph. D. in Linguistics