Em diferentes aplicações de Aprendizado de Máquina podemos estar interessados na minimização do valor esperado de certa função de perda. Para a resolução desse problema, Otimização estocástica e Sample Size Selection têm um papel importante. No presente trabalho se apresentam as análises teóricas de alguns algoritmos destas duas áreas, incluindo algumas variações que consideram redução da variância. Nos exemplos práticos pode-se observar a vantagem do método Stochastic Gradient Descent em relação ao tempo de processamento e memória, mas, considerando precisão da solução obtida juntamente com o custo de minimização, as metodologias de redução da variância obtêm as melhores soluções. Os algoritmos Dynamic Sample Size Gradient e Line Search with variable sample size selection apesar de obter soluções melhores que as de Stochastic Gradient Descent, a desvantagem se encontra no alto custo computacional deles. / In different Machine Learnings applications we can be interest in the minimization of the expected value of some loss function. For the resolution of this problem, Stochastic optimization and Sample size selection has an important role. In the present work, it is shown the theoretical analysis of some algorithms of these two areas, including some variations that considers variance reduction. In the practical examples we can observe the advantage of Stochastic Gradient Descent in relation to the processing time and memory, but considering accuracy of the solution obtained and the cost of minimization, the methodologies of variance reduction has the best solutions. In the algorithms Dynamic Sample Size Gradient and Line Search with variable sample size selection, despite of obtaining better solutions than Stochastic Gradient Descent, the disadvantage lies in their high computational cost.

Thesis advisor: Gail C. Kineke / This study characterizes the bed of the Connecticut River estuary in terms of grain size and bedforms, and relates these to river discharge, tidal currents, and sediment transport. Over four field excursions, sediment cores were collected, in addition to bathymetry surveys, and water column measurements. A three-dimensional circulation and sediment transport model calculated boundary shear stress over the same time. The bed of the estuary is composed mostly of sand, with small amounts of fine sediments. Deposition of fine sediments is limited by the landward extent of the salt intrusion. Large bedforms are oriented seaward. The critical shear stress for the median grain size is exceeded each tidal cycle. Bedload transport is dominantly seaward during high discharge conditions, but varies during low discharge. Bathymetry surveys from previous studies and this study show consistent bedform fields over 25 years. Bedforms observed in the field reflect typical conditions rather than extreme events. / Thesis (MS) — Boston College, 2015. / Submitted to: Boston College. Graduate School of Arts and Sciences. / Discipline: Earth and Environmental Sciences.

<div>New approaches in topology optimization and manufacturing techniques are generating multi-scale, physically realized mechanical components from advanced materials. Current optimization formulations do not consider the dependence of strength on feature size. By failing to account for the mechanical models of this behavior, sub-optimal structures are generated.</div><div><br></div><div>A currently available academic density-based topology optimization code is extended to incorporate strength constraints. A continuum theory of failure novel to the optimization field is implemented to account for both general yielding and fracture dominated failure. The fracture limit is then formulated in terms of well-established models of brittle and quasi-brittle size dependence. Additional models of size dependence based on assumed flaw sizes are considered using the theory of linear elastic fracture mechanics. To unify the optimized topology and the empirical geometric-scaling models used, a novel geometric measure of local size is proposed. This measure interprets the evolving density field using a consistent domain of support and maintains consistency with gradient-based methods of optimization. The geometric measure is evaluated using test-problems which consider a minimum compliance objective under geometric constraints.</div><div><br></div><div>The resulting optimized structures are presented for the geometric and size-dependent strength constrained formulations. The geometrically constrained results illustrate the flexibility and robustness of the proposed local size measure. The various models of size-dependent strength illustrate the impact and necessity of considering physical models of material within the topology optimization formulation. Results which exhibit clear "micro-structural" features and scale transitioning architectures are presented for limited multi-scale optimization studies.</div><div><br></div><div>An attempt at physical validation considering a single model of quasi-brittle material failure is made. Existing approaches for generating 3D volumetric meshes from image data are leveraged to yield CAD interpretations of optimized structures. Structures are then printed using a 3D printing PolyJet process with a previously established size-dependent material. Structures are destructively evaluated under displacement controlled load testing. The resulting tests indicate that the stress states in the structure fail to induce the expected size-dependent material characteristics. Furthermore, the testing results indicate the difficulty in properly accounting for boundary conditions in the topology optimization approach.</div>

As human populations increase and city borders grow, many animals have to modify their foraging behaviours to exploit evolutionarily novel urban food sources that could aid their survival. Neophobia, the fear of novelty, can lead to missed opportunities in these cases. Novelty is therefore expected to elicit different responses in urban and rural populations, a difference that has been frequently studied, but with mixed results. The main objective of my thesis was to study the novelty response of wild black-capped chickadees (Poecile atricapillus) in ecologically relevant conditions while controlling for individual characteristics and potential differences in foraging group size. I predicted that urban black-capped chickadees would be more likely to initially contact novelty than rural chickadees, and that subordinates and juveniles would be more likely to first contact novelty than dominants and adults, respectively. I ran replicated experiments using three novelty types (object, colour, or food) on six sites, during which I registered feeder choice of 71 tagged individuals. I found that urban chickadees showed less neophobia than their rural counterparts, the latter initially contacting the familiar feeder before approaching the novel feeder, while the former were equally likely to contact any feeder. There was no significant effect of an individual’s dominance, age or sex on its first choice of feeder, nor was there an effect of novelty type. Overall, my results suggest that urban chickadees exhibit less neophobia than their rural counterparts, because they have generally learned to tolerate novelty in their habitat and/or they have adapted to live in an environment that rewards low neophobia.

Most studies of plant hybridisation are concerned with documenting its occurrence in different plant groups. Many flowering plants are polyploids and seeds developed from crosses between individuals of different ploidies usually show abnormal features and often abort. The success or failure of interploidy crosses is very important to understanding the evolution of plants as well as to agriculture, but much remains to be learned about the nature of hybridisation barriers. Several mechanisms have been proposed to explain postzygotic barriers, including negative interactions between diverged sequences, global genome rearrangements, and widespread epigenetic reprogramming. Some recent advances in our understanding of the process of hybridisation are derived from different experimental studies on a series of A. thaliana ecotypes. Crosses between diploid (2x) and tetraploid (4x) individuals of the same ecotype can result in F1 lethality, and this dosage-sensitive incompatibility plays a major role in polyploidy speciation research. We have performed interploidy crosses between different diploid maternal A. thaliana ecotypes and tetraploid paternal Col-0 ecotype and identified a genetic variation in F1 lethality. We also found that maternal parents of some ecotypes such as Tsu-1 suppressed the F1 lethality caused by paternal-excess interploidy cross of Col-0 ecotype. A preliminary mapping exercise produced advanced backcross populations that are suitable for mapping maternal modifiers and for the identification of modifier genes. Furthermore, we studied the killer effect caused by Col-0 and identified three additive QTL that affect the rate of postzygotic lethality in F1 during interploidy crosses. This information will facilitate the identification of paternal genes that cause F1 lethality and contribute to reproductive isolation. The moa-1 (mosaic aneuploidy 1) mutant of A. thaliana was obtained in a screen of chemically (EMS) mutagenised seeds of Landsberg erecta (Ler). moa-1 has various phenotypic differences to wild type; the preliminary karyotype analysis showed that the cells of individual moa-1 mutant plants have a variable number of chromosomes (usually between 11-18). In contrast, the cells of wild type Arabidopsis plants and conventional aneuploids have a fixed number of chromosomes in each somatic cell. This data showed that all moa-1 plants have an abnormal number of chromosomes and thus they were termed as mosaic aneuploids.

"Experts in information theory have long been interested in the maximal size, A(n, d), of a binary error-correcting code of length n and minimum distance d, The problem of determining A(n, d) involves both the construction of good codes and the search for good upper bounds. For quite some time now, Delsarte's linear programming approach has been the dominant approach to obtaining the strongest general purpose upper bounds on the efficiency of error-correcting codes. From 1973 forward, the linear programming bound found many applications, but there were few significant theoretical advances until Schrijver proposed a new code upper bound via semidefinite programming in 2003. Using the Terwilliger algebra, a recently introduced extension of the Bose-Mesner algebra, Schrijver formulated a new SDP strengthening of the LP approach. In this project we look at the dual solutions of the semidefinite programming bound for binary error-correcting codes. We explore the combinatorial meaning of these variables for small n and d, such as n = 4 and d = 2. To obtain information like this, we wrote a computer program with both Matlab and CVX modules to get solution of our primal SDP formulation. Our program efficiently generates the primal solutions with corresponding constraints for any n and d. We also wrote a program in C++ to parse the output of the primal SDP problem, and another Matlab script to generate the dual SDP problem, which could be used in assigning combinatorial meaning to the values given in the dual optimal solution. Our code not only computes both the primal and dual optimal variable values, but allows the researcher to display them in meaningful ways and to explore their relationship and dependence on arameters. These values are expected to be useful for later study of the combinatorial meaning of such solutions."

Digitized by Kansas Correctional Industries

The superfamily Sauropodomorpha comprised some of the largest terrestrial herbivorous vertebrates that have ever walked on this planet. The study of these fascinating beings has been at the forefront of palaeontological research throughout the years. Especially in the last 20 years, due to the advent of modern techniques and technologies, significant steps have been made towards our understanding of their physiology. One of the sauropodomorphs’ central morphological characteristics is the expression of postcranial skeletal pneumaticity (PSP), a condition that is also expressed in the other superfamily of the order Saurischia, the Theropoda, as well as in their extant relatives, the Avians. This condition remodels the vertebrae and, occasionally the girdles and appendicular elements, producing perforations, depressions, excavations, cavities and internal chambers. This is a result of bone invasion and resorption from the development of the lung-air sac diverticula of the respiratory system. The various forms of this expression have been addressed across all subfamilies of sauropodomorphs mostly from an evolutionary aspect and, recently, from a developmental scope too. The resulting hypotheses state that PSP may have acted as a mechanism for weight reduction, allowing for sauropods to attain large sizes without having to suffer analogous gravitational constraints from equally attained masses. Any possible associations, though, between the expression of pneumaticity and body size have not been put under test. Through this study, a method of quantification and categorisation of PSP, and therefore, classification of the sauropodomorphs which express it, is created from the data retrieved from 61 taxa across all subfamilies, permitting us not only to trace any correlation between PSP and metric size data (body mass and body length) but also to visualise the evolution of PSP throughout Sauropodomorpha. This classification scheme from highest to lowest expression of PSP, ‘Alpha’, ‘Beta’, ‘Gamma’, ‘Delta’, and ‘Epsilon’ stems from the numerical estimate of pneumaticity in terms of a percentage, called the Pneumaticity Degree Index (PDI%). The revised scheme, Pneumaticity Degree Index (PDI%), takes account of the number of vertebral elements that are pneumatized in a single vertebra, the nature of pneumaticity traits, as well as the intensity of pneumatization in different body regions of interest (e.g. vertebral column), resulting in an integral and comprehensive measure of PSP. The proposed method ranks each pneumaticity trait with a value from 1 to 5 with 1 (100%) representing the most invasive unambiguous trait (e.g. camellae) and 5 (~ 10%) the least invasive and most ambiguous trait (i.e. fossa). By adding and dividing by their number all of the observed traits of every available vertebra of a region we retrieve a decimal numeric outcome and this outcome is translated to a percentage. The total average pneumatization of any taxon is calculated by retrieving the total average of the pneumaticity from all available body regions. In this study, only vertebral and pelvic elements were used for the retrieval of pneumaticity data since they are the most frequently exhibiting pneumatic traits in comparison to pectoral and appendicular elements. Results show that the total average PDI% range of basal Sauropodomorpha is 0%-59%, of the non-neosauropod Eusauropoda is 23%-90%, of Diplodocoidea is 45%-73%, of Macronaria is 45%-92% and of Somphospondyli is 46%-94%. The most pneumatized vertebral landmarks are the centrum, neural arch and neural spine. The vertebral regions most commonly pneumatized are the cervicals and dorsals. No genus or subfamily of sauropodomorphs ever reaches 100% pneumatization in all vertebral regions. Furthermore, PSP is not always positively correlated with mass, rather its expression is mostly correlated with length. Taxa with low or high masses may exhibit either low or high PDI’s. In addition, increasing progression of the extent and expression of PSP occurs mostly on a subfamilial level and less throughout the entire lineage. Finally, modelling calculations result in an Ornstein-Ulenbeck with an early burst progress of the expression of pneumatisation in sauropodomorphs. After that, the expression proceeds relatively steadily throughout the entire superfamily. The biotic causes of PSP are still unclear, as it could be an artefact of inheritance and genetic drift throughout speciation events. The pneumatization degree index (PDI%) is a means of quantifying and categorising pneumatization in any archosaurian taxon that is faster and cheaper, though less accurate, than CT scanning. It is more precise than the Pneumaticity Index since it integrates a wider selection of the qualities and osteological characteristics of pneumaticity we want to measure, provided that the bone under study is at least 50% free of matrix.

Em diferentes aplicações de Aprendizado de Máquina podemos estar interessados na minimização do valor esperado de certa função de perda. Para a resolução desse problema, Otimização estocástica e Sample Size Selection têm um papel importante. No presente trabalho se apresentam as análises teóricas de alguns algoritmos destas duas áreas, incluindo algumas variações que consideram redução da variância. Nos exemplos práticos pode-se observar a vantagem do método Stochastic Gradient Descent em relação ao tempo de processamento e memória, mas, considerando precisão da solução obtida juntamente com o custo de minimização, as metodologias de redução da variância obtêm as melhores soluções. Os algoritmos Dynamic Sample Size Gradient e Line Search with variable sample size selection apesar de obter soluções melhores que as de Stochastic Gradient Descent, a desvantagem se encontra no alto custo computacional deles. / In different Machine Learnings applications we can be interest in the minimization of the expected value of some loss function. For the resolution of this problem, Stochastic optimization and Sample size selection has an important role. In the present work, it is shown the theoretical analysis of some algorithms of these two areas, including some variations that considers variance reduction. In the practical examples we can observe the advantage of Stochastic Gradient Descent in relation to the processing time and memory, but considering accuracy of the solution obtained and the cost of minimization, the methodologies of variance reduction has the best solutions. In the algorithms Dynamic Sample Size Gradient and Line Search with variable sample size selection, despite of obtaining better solutions than Stochastic Gradient Descent, the disadvantage lies in their high computational cost.

Post-copulatory sexual selection is a strong evolutionary force, affecting morphological and behavioural traits in males and females in species with polyandrous mating systems. Many insects are subject to sperm competition; sperm from rival males compete to fertilise ova. Since sperm are finite, males should allocate them economically, tailoring ejaculate allocation to suit the reproductive potential of individual matings. Theory suggests when sperm competition risk is high, males should increase sperm numbers to achieve greater reproductive success than their rivals, but evidence of this expected fitness consequence of ejaculate allocation is largely lacking. In this thesis, I use Callosobruchus maculatus beetles to investigate the causes of ejaculate allocation patterns, and to examine whether ejaculate allocation does affect male reproductive success. In Chapter 3, I investigate the effect of rival male presence on ejaculate size and find that, while males grouped with rivals as adults produce bigger ejaculates, their increased effort unexpectedly does not lead to increased reproductive success. In Chapter 4, I examine whether larval conditions also affect ejaculate size, and find that, contrary to sperm competition theory, males reared under dense conditions produce smaller ejaculates than those reared solitarily, and that male reproductive success is consequently elevated in males reared at low larval densities compared to those reared at high densities. In Chapter 5, I then demonstrate that ejaculates produced by low density males contain more sperm than ejaculates produced by high density males, suggesting males do not respond to sperm competition level represented by larval density, but instead suffer resource limitation when reared at high density. In Chapter 6, I investigate the effects of water provision on ejaculate size, and find that males given water produce larger ejaculates, and females given water receive smaller ejaculates. Finally, I link my findings with those of other studies, and suggest my most important result is that plasticity of ejaculate allocation cannot be assumed to be an adaptive behaviour; studies directly measuring the fitness effects of male ejaculate allocation are needed, even when observed patterns conform to theory.