L'utilisation d'aspects aléatoires a contribué de façon majeure aux dernières avancées dans le domaine de l'optimisation numérique; cela est dû en partie à la recrudescence de problèmes issus de l'apprentissage automatique (machine learning). Dans un tel contexte, les algorithmes classiques d'optimisation non linéaire, reposant sur des principes déterministes, se révèlent en effet bien moins performants que des variantes incorporant de l'aléatoire. Le coût de ces dernières est souvent inférieur à celui de leurs équivalents déterministes; en revanche, il peut s'avérer difficile de maintenir les propriétés théoriques d'un algorithme déterministe lorsque de l'aléatoire y est introduit. Effectuer une analyse de complexité d'une telle méthode est un procédé très répandu dans ce contexte. Cette technique permet déstimer la vitesse de convergence du schéma considéré et par là même d'établir une forme de convergence de celui-ci. Les récents travaux sur ce sujet, en particulier pour des problèmes d'optimisation non convexes, ont également contribué au développement de ces aspects dans le cadre déterministe, ceux-ci apportant en effet un éclairage nouveau sur le comportement des algorithmes. Dans cette thèse, on s'intéresse à l'amélioration pratique d'algorithmes d'optimisation sans dérivées à travers l'introduction d'aléatoire, ainsi qu'à l'impact numérique des analyses de complexité. L'étude se concentre essentiellement sur les méthodes de recherche directe, qui comptent parmi les principales catégories d'algorithmes sans dérivées; cependant, l'analyse sous-jacente est applicable à un large éventail de ces classes de méthodes. On propose des variantes probabilistes des propriétés requises pour assurer la convergence des algorithmes étudiés, en mettant en avant le gain en efficacité induit par ces variantes: un tel gain séxplique principalement par leur coût très faible en évaluations de fonction. Le cadre de base de notre analyse est celui de méthodes convergentes au premier ordre, que nous appliquons à des problèmes sans ou avec contraintes linéaires. Les bonnes performances obtenues dans ce contexte nous incitent par la suite à prendre en compte des aspects d'ordre deux. A partir des propriétés de complexité des algorithmes sans dérivées, on développe de nouvelles méthodes qui exploitent de l'information du second ordre. L'analyse de ces procédures peut être réalisée sur un plan déterministe ou probabiliste: la deuxième solution nous permet d'étudier de nouveaux aspects aléatoires ainsi que leurs conséquences sur l'éfficacité et la robustesse des algorithmes considérés. / Randomization has had a major impact on the latest developments in the field of numerical optimization, partly due to the outbreak of machine learning applications. In this increasingly popular context, classical nonlinear programming algorithms have indeed been outperformed by variants relying on randomness. The cost of these variants is usually lower than for the traditional schemes, however theoretical guarantees may not be straightforward to carry out from the deterministic to the randomized setting. Complexity analysis is a useful tool in the latter case, as it helps in providing estimates on the convergence speed of a given scheme, which implies some form of convergence. Such a technique has also gained attention from the deterministic optimization community thanks to recent findings in the nonconvex case, as it brings supplementary indicators on the behavior of an algorithm. In this thesis, we investigate the practical enhancement of deterministic optimization algorithms through the introduction of random elements within those frameworks, as well as the numerical impact of their complexity results. We focus on direct-search methods, one of the main classes of derivative-free algorithms, yet our analysis applies to a wide range of derivative-free methods. We propose probabilistic variants on classical properties required to ensure convergence of the studied methods, then enlighten their practical efficiency induced by their lower consumption of function evaluations. Firstorder concerns form the basis of our analysis, which we apply to address unconstrained and linearly-constrained problems. The observed gains incite us to additionally take second-order considerations into account. Using complexity properties of derivative-free schemes, we develop several frameworks in which information of order two is exploited. Both a deterministic and a probabilistic analysis can be performed on these schemes. The latter is an opportunity to introduce supplementary probabilistic properties, together with their impact on numerical efficiency and robustness.

In this thesis, we investigate methods of finding a local minimum for unconstrained problems of non-convex functions with n variables, by following the solution curve of a system of ordinary differential equations. The motivation for this was the fact that existing methods (e.g. those based on Newton methods with line search) sometimes terminate at a non-stationary point when applied to functions f(x) that do not a have positive-definite Hessian (i.e. ∇²f → 0) for all x. Even when methods terminate at a stationary point it could be a saddle or maximum rather than a minimum. The only method which makes intuitive sense in non-convex region is the trust region approach where we seek a step which minimises a quadratic model subject to a restriction on the two-norm of the step size. This gives a well-defined search direction but at the expense of a costly evaluation. The algorithms derived in this thesis are gradient based methods which require systems of equations to be solved at each step but which do not use a line search in the usual sense. Progress along the Continuous Steepest Descent Path (CSDP) is governed both by the decrease in the function value and measures of accuracy of a local quadratic model. Numerical results on specially constructed test problems and a number of standard test problems from CUTEr [38] show that the approaches we have considered are more promising when compared with routines in the optimization tool box of MATLAB [46], namely the trust region method and the quasi-Newton method. In particular, they perform well in comparison with the, superficially similar, gradient-flow method proposed by Behrman [7].

Orientador: Maria Aparecida Diniz Ehrhardt / Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Matemática Estatística e Computação / Made available in DSpace on 2018-08-22T06:10:21Z (GMT). No. of bitstreams: 1 Ferreira_DeiseGoncalves_M.pdf: 2631020 bytes, checksum: 45eb84901394375843735b1fdef599ad (MD5) Previous issue date: 2013 / Resumo: Neste trabalho voltamos nossa atenção para métodos de otimização que não fazem uso de derivadas. Dentre esses, estamos interessadas em um método de busca padrão para minimização de funções com restrições lineares. Abordamos um algoritmo proposto por Lewis e Torczon, cuja ideia geral é que o padrão deve conter direções de busca ao longo das quais iterações factíveis sejam determinadas. O algoritmo possui resultados de convergência global. Realizamos sua implementação computacional, e propomos novas estratégias de busca e atualização do tamanho do passo, além de um novo padrão de direções de busca. Realizamos testes numéricos, de modo a analisar o desempenho das estratégias propostas e comparar o desempenho do padrão de direções que introduzimos com o proposto por Lewis e Torczon / Abstract: In this work, our interest lies on derivative-free optimization methods. Among these, our aim is to study a pattern search method for linearly constrained minimization. We studied an algorithm proposed by Lewis and Torsion, whose general idea is that the pattern must contain search directions in which feasible iterations must be computed. The algorithm has global convergence results. We accomplished its computational implementation and we propose new strategies of search and updating rule for the step-length control parameter. We also propose a new pattern of search directions. We accomplished numerical experiments in order to analyze the performance of our proposals and also to compare the performance of our pattern with the one proposed by Lewis and Torczon / Mestrado / Matematica Aplicada / Mestra em Matemática Aplicada

Student Number : 0214677F - PhD thesis - School of Camputational and Applied Mathematics - Faculty of Science / Many real-life problems are formulated as global optimization problems with continuous variables. These problems are in most cases nonsmooth, nonconvex and often simulation based, making gradient based methods impossible to be used to solve them. Therefore, ef#2;cient, reliable and derivative-free global optimization methods for solving such problems are needed. In this thesis, we focus on improving the ef#2;ciency and reliability of some global optimization methods. In particular, we concentrate on improving some population set-based methods for unconstrained global optimization, mainly through hybridization. Hybridization has widely been recognized to be one of the most attractive areas of unconstrained global optimization. Experiments have shown that through hybridization, new methods that inherit the strength of the original elements but not their weakness can be formed. We suggest a number of new hybridized population set-based methods based on differential evolution (de), controlled random search (crs2) and real coded genetic algorithm (ga). We propose #2;ve new versions of de. In the #2;rst version, we introduce a localization, called random localization, in the mutation phase of de. In the second version, we propose a localization in the acceptance phase of de. In the third version, we form a de hybrid algorithm by probabilistically combining the point generation scheme of crs2 with that of de in the de algorithm. The fourth and #2;fth versions are also de hybrids. These versions hybridize the mutation of de with the point generation rule of the electromagnetism-like (em) algorithm. We also propose #2;ve new versions of crs2. The #2;rst version modi#2;es the point generation scheme of crs2 by introducing a local mutation technique. In the second and third modi#2;cations, we probabilistically combine the point generation scheme of crs2 with the linear interpolation scheme of a trust-region based method. The fourth version is a crs hybrid that probabilistically combines the quadratic interpolation scheme with the linear interpolation scheme in crs2. In the #2;fth version, we form a crs2 hybrid algorithm by probabilistically combining the point generation scheme of crs2 with that of de in the crs2 algorithm. Finally, we propose #2;ve new versions of the real coded genetic algorithm (ga) with arithmetic crossover. In the #2;rst version of ga, we introduce a local technique. We propose, in the second version, an integrated crossover rule that generates two children at a time using two different crossover rules. We introduce a local technique in the second version to obtain the third version. The fourth and #2;fth versions are based on the probabilistic adaptation of crossover rules. The ef#2;ciency and reliability of the new methods are evaluated through numerical experiments using a large test suite of both simple and dif#2;cult problems from the literature. Results indicate that the new hybrids are much better than their original counterparts both in reliability and ef#2;ciency. Therefore, the new hybrids proposed in this study offer an alternative to many currently available stochastic algorithms for solving global optimization problems in which the gradient information is not readily available.

<p>Within telecommunication and routing of traffic in IP-networks a protocol named“Open Shortest Path First” (OSPF) is widely used. This means that a server dealswith the routing over a network with given weights by calculating shortest paths touse for routing. If we assume that a desired traffic pattern is given the problem isto find out if it is possible to set the weights so that the desired traffic pattern is apart of a shortest path graph. In this thesis we assume that it is a unique shortestpath. To search for weights that solve the problem leads to a complex LP-model. Analternative is to search in the LP-dual under certain restrictions. These solutions tothe LP-dual are called conflicts and a conflict means that there exists no weights sothat the desired traffic pattern is obtained. The goal of this thesis is to study, classifyand search for conflicts. An algorithm has been developed that finds some kind ofconflicts in polynomial time with respect to the size of the graph.</p> / <p>Inom telekommunikation och ruttning av datatrafik i IP-nätverk så används oftaett protokoll som kallas “Open Shortest Path First” (OSPF). Det innebär att enserver sköter ruttningen över ett nätverk genom att utifrån givna bågkostnaderberäkna billigaste vägar som används för ruttningen. Frågeställningen utgårfrån att vi har ett önskat ruttningsschema och vi vill ta reda på om det gåratt sätta bågkostnader så att det önskade ruttningsschemat ingår i en billigasteväg-graf. I det här examensarbetet splittas inte trafik utan varje billigaste vägär unik mellan två noder. Att söka efter bågkostnader som löser problemet geren krävande LP-modell och ett alternativ är att utgå från LP-dualen undervissa restriktioner. Dessa lösningar till LP-dualen benämns konflikter och enkonflikt motsvarar att det inte finns några bågkostnader så att det önskaderuttningsschemat fås. Målet med examensarbetet är att studera, klassificeraoch söka efter konflikter. En algoritm har tagits fram som hittar vissa typer avsådana konflikter i polynomiell tid, sett till storleken på grafen.</p>

Within telecommunication and routing of traffic in IP-networks a protocol named“Open Shortest Path First” (OSPF) is widely used. This means that a server dealswith the routing over a network with given weights by calculating shortest paths touse for routing. If we assume that a desired traffic pattern is given the problem isto find out if it is possible to set the weights so that the desired traffic pattern is apart of a shortest path graph. In this thesis we assume that it is a unique shortestpath. To search for weights that solve the problem leads to a complex LP-model. Analternative is to search in the LP-dual under certain restrictions. These solutions tothe LP-dual are called conflicts and a conflict means that there exists no weights sothat the desired traffic pattern is obtained. The goal of this thesis is to study, classifyand search for conflicts. An algorithm has been developed that finds some kind ofconflicts in polynomial time with respect to the size of the graph. / Inom telekommunikation och ruttning av datatrafik i IP-nätverk så används oftaett protokoll som kallas “Open Shortest Path First” (OSPF). Det innebär att enserver sköter ruttningen över ett nätverk genom att utifrån givna bågkostnaderberäkna billigaste vägar som används för ruttningen. Frågeställningen utgårfrån att vi har ett önskat ruttningsschema och vi vill ta reda på om det gåratt sätta bågkostnader så att det önskade ruttningsschemat ingår i en billigasteväg-graf. I det här examensarbetet splittas inte trafik utan varje billigaste vägär unik mellan två noder. Att söka efter bågkostnader som löser problemet geren krävande LP-modell och ett alternativ är att utgå från LP-dualen undervissa restriktioner. Dessa lösningar till LP-dualen benämns konflikter och enkonflikt motsvarar att det inte finns några bågkostnader så att det önskaderuttningsschemat fås. Målet med examensarbetet är att studera, klassificeraoch söka efter konflikter. En algoritm har tagits fram som hittar vissa typer avsådana konflikter i polynomiell tid, sett till storleken på grafen.

Orientador: Maria Aparecida Diniz Ehrhardt / Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Matemática Estatística e Computação Científica / Made available in DSpace on 2018-08-22T15:48:00Z (GMT). No. of bitstreams: 1 Cervelin_BrunoHenrique_M.pdf: 1935510 bytes, checksum: 91d17dd60bdd280c9eddd301cb3d2c24 (MD5) Previous issue date: 2013 / Resumo: O objetivo deste trabalho é apresentar alguns métodos de minimização irrestrita sem derivadas, tais como, Nelder-Mead, busca padrão e SID-PSM, assim como compará-los. Ainda pretendemos apresentar o problema de otimização de parâmetros de algoritmos, e aplicar o método SID-PSM de modo a encontrar parâmetros ótimos para o próprio método SID-PSM em relação ao número de avaliações de função que o método realiza. Os experimentos numéricos realizados mostram que o SID-PSM _e mais robusto e mais eficiente que os métodos clássicos sem derivadas (busca padrão e Nelder-Mead). Outros experimentos nos mostram o potencial do problema de otimização de parâmetros de algoritmos em melhorar tanto a eficiência quanto a robustez dos métodos / Abstract: The aim of this paper is to present some derivative-free methods for unconstrained minimization problems, such as Nelder-Mead, pattern search and SID-PSM, and compare them. We also intend to present the problem of optimal algorithmic parameters, and apply the method SID-PSM in order to find optimal parameters for the method SID-PSM itself in relation to the number of function evaluations performed by the method. The numerical experiments performed show that the SID-PSM is more robust and more efficient than the classical derivative-free methods (pattern search and Nelder-Mead). Other experiments show us the potential of the problem of optimal algorithmic parameters to improve both the efficiency and the robustness of the methods / Mestrado / Matematica Aplicada / Mestre em Matemática Aplicada

The term ‘Object Shift’ means that a weak object pronoun has been raised to a position to the left of a sentence adverbial. Primarily, there are two positions to which a weak object pronoun can be raised; one of them is called Long Object Shift. In this paper, I primarily deal with finite verb frequencies in sentences with Long Object Shift in Swedish. Also, in order to further define the verbs in these sentences I discuss verb semantic and syntactic valence in relation to the shifted object. Furthermore, I discuss the search methods which have generated sentences with Long Object Shift in written sources in Swedish; an equivalent study has never been done before.   The most salient results of my study indicate that: verbs with physical denotation, and specifically vänta, möta and slå, frequently recur in sentences with Long Object Shift in written sources in Swedish; monotransitive verbs appear more often than ditransitive ones; shifted objects are most often assigned the semantic role of experiencer in the studied sentences; the search methods show divergent results, which indicates the significance of sentence adverbial initiating sentences with Long Object Shift. Lastly, the results show that semantic analyses, terminology and search methods need to be extended in order to further define the finite verb in sentences with Long Object Shift in Swedish.

Les récents mouvements sociaux qui ont secoué certains pays arabes notamment l'Algérie ont montré la fragilité de la situation sur le marché du travail dans ces pays en particulier pour les jeunes. Le premier chapitre porte sur l'analyse de la situation du marché du travail en Algérie sur une longue période. Ces dix dernières années la situation de l'emploi s'est dégradée en termes de sécurité de l'emploi en particulier pour les jeunes. La situation reste difficile, d'autant plus que la situation démographique actuelle envisage de fortes pressions sur le marché du travail dans le futur. Dans le deuxième chapitre, nous analysons les stratégies de recherche d'emploi notamment par le biais des méthodes informelles. Nos résultats montrent de fortes spécificités liées au genre dans les stratégies de recherche. Dans l'ensemble, les réseaux sociaux apparaissent comme étant une méthode efficace pour trouver un emploi en Algérie mais pas pour des emplois qualifiés. Le troisième chapitre vise à étudier l'architecture des systèmes d'emploi en Algérie. Il ressort que le modèle qui explique le mieux le comportement des hommes et des femmes est un modèle séquentiel à deux niveaux : participation vs non participation et ensuite le choix d'un segment sur le marché du travail. Dans le quatrième chapitre, nous analysons les déterminants du choix du secteur et les salaires pour les hommes et les femmes dans différents secteurs d'activités en Algérie. Nos résultats montrent que les femmes sont moins bien rémunérées par rapport aux hommes dans les différents segments. Le secteur public est celui qui protège le plus les femmes de la discrimination salariale. / Recent social movements that swept some Arab countries including Algeria showed the fragility of the situation on the labor market in these countries in particular for young people. The first chapter deals with the analysis of the situation of the labor market in Algeria over a long period. Over the past decade the employment situation has deteriorated in terms of job security in particular for young people. The situation remains difficult, especially as the current demographic situation considering pressures on the labor market in the future. In the second chapter, we analyze the strategies for job search including through informal methods. Our results show strong gender specificities in search strategies. Overall, social networks appear to be an effective method to find a job in Algeria but not for skilled jobs. The third chapter aims to study the architecture of employment systems in Algeria. It appears that the model that best explains the behavior of men and women is a sequential two-level model: participation vs. non participation and then choosing a segment on the labor market. In the fourth chapter, we analyze the determinants of the choice of sector and wages for men and women in different sectors in Algeria. Our results show that women are paid less compared to men in the various segments. The public sector is the one that best protects women from wage discrimination.

<p>The problem under consideration is an unconstrained optimization&nbsp;problem with the objective function in the form of mathematical ex-pectation. The expectation is with respect to the random variable that represents the uncertainty. Therefore, the objective &nbsp;function is in fact deterministic. However, nding the analytical form of that objective function can be very dicult or even impossible. This is the reason why the sample average approximation is often used. In order to obtain reasonable good approximation of the objective function, we have to use relatively large sample size. We assume that the sample is generated at the beginning of the optimization process and therefore we can consider this sample average objective function as the deterministic one. However, applying some deterministic method on that sample average function from the start can be very costly. The number of evaluations of the function under expectation is a common way of measuring the cost of an algorithm. Therefore, methods that vary the sample size throughout the optimization process are developed. Most of them are trying to determine the optimal dynamics of increasing the sample size.</p><p>The main goal of this thesis is to develop the clas of methods that&nbsp;can decrease the cost of an algorithm by decreasing the number of&nbsp;function evaluations. The idea is to decrease the sample size whenever&nbsp;it seems to be reasonable - roughly speaking, we do not want to impose&nbsp;a large precision, i.e. a large sample size when we are far away from the&nbsp;solution we search for. The detailed description of the new methods&nbsp;<br />is presented in Chapter 4 together with the convergence analysis. It&nbsp;is shown that the approximate solution is of the same quality as the&nbsp;one obtained by dealing with the full sample from the start.</p><p>Another important characteristic of the methods that are proposed&nbsp;here is the line search technique which is used for obtaining the sub-sequent iterates. The idea is to nd a suitable direction and to search&nbsp;along it until we obtain a sucient decrease in the &nbsp;function value. The&nbsp;sucient decrease is determined throughout the line search rule. In&nbsp;Chapter 4, that rule is supposed to be monotone, i.e. we are imposing&nbsp;strict decrease of the function value. In order to decrease the cost of&nbsp;the algorithm even more and to enlarge the set of suitable search directions, we use nonmonotone line search rules in Chapter 5. Within that chapter, these rules are modied to t the variable sample size framework. Moreover, the conditions for the global convergence and the R-linear rate are presented.&nbsp;</p><p>In Chapter 6, numerical results are presented. The test problems&nbsp;are various - some of them are academic and some of them are real&nbsp;world problems. The academic problems are here to give us more&nbsp;insight into the behavior of the algorithms. On the other hand, data&nbsp;that comes from the real world problems are here to test the real&nbsp;applicability of the proposed algorithms. In the rst part of that&nbsp;chapter, the focus is on the variable sample size techniques. Different&nbsp;implementations of the proposed algorithm are compared to each other&nbsp;and to the other sample schemes as well. The second part is mostly&nbsp;devoted to the comparison of the various line search rules combined&nbsp;with dierent search directions in the variable sample size framework.&nbsp;The overall numerical results show that using the variable sample size&nbsp;can improve the performance of the algorithms signicantly, especially&nbsp;when the nonmonotone line search rules are used.</p><p>The rst chapter of this thesis provides the background material&nbsp;for the subsequent chapters. In Chapter 2, basics of the nonlinear&nbsp;optimization are presented and the focus is on the line search, while&nbsp;Chapter 3 deals with the stochastic framework. These chapters are&nbsp;here to provide the review of the relevant known results, while the&nbsp;rest of the thesis represents the original contribution.&nbsp;</p> / <p>U okviru ove teze posmatra se problem optimizacije bez ograničenja pri čcemu je funkcija cilja u formi matematičkog očekivanja. Očekivanje se odnosi na slučajnu promenljivu koja predstavlja neizvesnost. Zbog toga je funkcija cilja, u stvari, deterministička veličina. Ipak, odredjivanje analitičkog oblika te funkcije cilja može biti vrlo komplikovano pa čak i nemoguće. Zbog toga se za aproksimaciju često koristi uzoračko očcekivanje. Da bi se postigla dobra aproksimacija, obično je neophodan obiman uzorak. Ako pretpostavimo da se uzorak realizuje pre početka procesa optimizacije, možemo posmatrati uzoračko očekivanje kao determinističku funkciju. Medjutim, primena nekog od determinističkih metoda direktno na tu funkciju&nbsp; moze biti veoma skupa jer evaluacija funkcije pod ocekivanjem često predstavlja veliki tro&scaron;ak i uobičajeno je da se ukupan tro&scaron;ak optimizacije meri po broju izračcunavanja funkcije pod očekivanjem. Zbog toga su razvijeni metodi sa promenljivom veličinom uzorka. Većcina njih je bazirana na odredjivanju optimalne dinamike uvećanja uzorka.</p><p>Glavni cilj ove teze je razvoj algoritma koji, kroz smanjenje broja izračcunavanja funkcije, smanjuje ukupne tro&scaron;skove optimizacije. Ideja je da se veličina uzorka smanji kad god je to moguće. Grubo rečeno, izbegava se koriscenje velike preciznosti&nbsp; (velikog uzorka) kada smo daleko od re&scaron;senja. U čcetvrtom poglavlju ove teze opisana je nova klasa metoda i predstavljena je analiza konvergencije. Dokazano je da je aproksimacija re&scaron;enja koju dobijamo bar toliko dobra koliko i za metod koji radi sa celim uzorkom sve vreme.</p><p>Jo&scaron; jedna bitna karakteristika metoda koji su ovde razmatrani je primena linijskog pretražzivanja u cilju odredjivanja naredne iteracije. Osnovna ideja je da se nadje odgovarajući pravac i da se duž njega vr&scaron;si pretraga za dužzinom koraka koja će dovoljno smanjiti vrednost funkcije. Dovoljno smanjenje je odredjeno pravilom linijskog pretraživanja. U čcetvrtom poglavlju to pravilo je monotono &scaron;to znači da zahtevamo striktno smanjenje vrednosti funkcije. U cilju jos većeg smanjenja tro&scaron;kova optimizacije kao i pro&scaron;irenja skupa pogodnih pravaca, u petom poglavlju koristimo nemonotona pravila linijskog pretraživanja koja su modifikovana zbog promenljive velicine uzorka. Takodje, razmatrani su uslovi za globalnu konvergenciju i R-linearnu brzinu konvergencije.</p><p>Numerički rezultati su predstavljeni u &scaron;estom poglavlju. Test problemi su razliciti - neki od njih su akademski, a neki su realni. Akademski problemi su tu da nam daju bolji uvid u pona&scaron;anje algoritama. Sa druge strane, podaci koji poticu od stvarnih problema služe kao pravi test za primenljivost pomenutih algoritama. U prvom delu tog poglavlja akcenat je na načinu ažuriranja veličine uzorka. Različite varijante metoda koji su ovde predloženi porede se medjusobno kao i sa drugim &scaron;emama za ažuriranje veličine uzorka. Drugi deo poglavlja pretežno je posvećen poredjenju različitih pravila linijskog pretraživanja sa različitim pravcima pretraživanja u okviru promenljive veličine uzorka. Uzimajuci sve postignute rezultate u obzir dolazi se do zaključcka da variranje veličine uzorka može značajno popraviti učinak algoritma, posebno ako se koriste nemonotone metode linijskog pretraživanja.</p><p>U prvom poglavlju ove teze opisana je motivacija kao i osnovni pojmovi potrebni za praćenje preostalih poglavlja. U drugom poglavlju je iznet pregled osnova nelinearne optimizacije sa akcentom na metode linijskog pretraživanja, dok su u trećem poglavlju predstavljene osnove stohastičke optimizacije. Pomenuta poglavlja su tu radi pregleda dosada&scaron;njih relevantnih rezultata dok je originalni doprinos ove teze predstavljen u poglavljima 4-6.</p>