Browsing by Author "Johnson, Nicholas"
Now showing 1 - 5 of 5
Results Per Page
Sort Options
Item Flexural Performance Of Reinforced Concrete Beams Fabricated With Ground Glass And Fly Ash(2020-05) Johnson, NicholasConcrete is an essential construction material; however, the production of cement (a component of concrete) produces large quantities of carbon dioxide. These large quantities of anthropogenic carbon dioxide (along with other reasons) encouraged a shift toward the use of supplementary cementitious materials (SCMs) such as fly ash or granulated blast furnace slag, which are byproducts of other industries. Ground waste glass is another possible SCM that, at the proper particle size and depending on the chemical composition of the glass, can produce concrete with comparable or higher strength than concrete made without SCMs. However, there is a lack of research related to using glass as an SCM in structural concrete with steel reinforcement. In this study, nine concrete mixtures were created using various combinations of two fly ashes and three different types of ground waste glass and were compared to a control batch of concrete without SCMs. Three 6 ft beam specimens were fabricated from each of these ten mixtures (30 beams total) and the flexural behavior was investigated. Results from the 90-day flexural testing demonstrated that three out of the seven mixtures containing ground glass had higher average flexural strengths than the control beams (between 0.5-4% higher). The remaining beams made with glass had approximately 2-5% less flexural strength compared to control beams. However, there was no statistically significant difference in flexural strength between each of the nine SCM mixtures and the control beams without SCMs. Additionally, the beams made with ground glass had a lower displacement at failure in all cases when compared to the control beams.Item Intraspecific Salt Tolerance Variation in Nicotiana tabacum Pollen Germination and Pollen Tube Growth(2019-08) Johnson, NicholasSalt stress affects 20% of global cropland. Excessive salt inhibits pollen germination and pollen tube growth (PTG), two crucial aspects of angiosperm ontogenesis. Pollen tolerance to salt and how tolerances vary intraspecifically are important factors of the sexual reproduction and evolution of flowering plants. Additionally, pollen response to salt is an effective screening method for whole plant salt tolerance. While the general field of plant salt tolerance is well studied, little research has been done on intraspecific variation in salt effect on pollen germination and PTG. The objective of this study was to investigate the effect of super-optimal levels of sodium chloride (NaCl) on pollen germination and PTG in vitro in 11 diverse genotypes of Nicotiana tabacum. NaCl tolerance was evaluated in three concentrations in N. tabacum pollen tube growth medium (TPTGM), where tolerance was observed in terms of PTG and pollen germination. ‘Mpeskq’ is NaCl tolerant relative to 10 other genotypes of N. tabacum. ‘Mpsekq’ pollen germination percentage was reduced by 5.8% and PTG was reduced by 34.6% from control TPTGM to 50mM NaCl TPTGM treatments. ‘Aparecido’ was the least NaCl tolerant, with a reduction of 62.6% PTG and 93.5% pollen germination between these same treatments. A strong positive correlation between PTG NaCl tolerance and pollen germination NaCl tolerance was also discovered (r = 0.74, p-value = 7.54e^(-7)). These findings show there is genetic variation for NaCl tolerance among diverse genotypes of N. tabacum. Future research should confirm this work and identify the genes responsible for NaCl resistance among genotypes such as ‘Mpsekq’ for use in N. tabacum or other crops.Item Stock Portfolio Selection Using Two-tiered Lazy Updates(2014-04-16) Cook, Alexander; Johnson, Nicholas; Banerjee, ArindamPeople make and lose vast sums of money every day on stock exchanges around the world. This research focused on developing a computer algorithm to build profitable portfolios, while taking into account transaction costs associated with trading stocks. The theory behind our algorithm is based on a subset of Machine Learning called Online Learning. Online Learning makes updated decisions as new information is provided. For our case, a new decision is made each day on what stocks to buy/sell based on transaction costs and the previous day’s stock performance. The Lazy Update part of our algorithm seeks to minimize the quantity of trading, since this leads to transaction costs being incurred. Our algorithm builds on prior work and dynamically learns which sectors to invest in and takes into account risk, which has not been considered before in the literature. Our Online Lazy Updates algorithm runs at a low level on choosing stocks within a sector, and at a high level on choosing the best sectors to invest in. We successfully establish our ability to be profitable with transaction costs on real-world datasets.Item Structured Online Learning with Full and Bandit Information(2016-09) Johnson, NicholasNumerous problems require algorithms to repeatedly interact with the environment which may include humans, physical obstacles such as doors or walls, biological processes, or even other algorithms. One popular application domain where such repeated interaction is necessary is in social media. Every time a user uses a social media application, an algorithm must make a series of decisions on what is shown ranging from news content to friend recommendations to trending topics. After which, the user provides feedback frequently in the form of a click or no click. The algorithm must use such feedback to learn the user's likes and dislikes. Similar scenarios play out in medical treatments, autonomous robot exploration, online advertising, and algorithmic trading. In such applications, users often have high expectations of the algorithm such as immediately showing relevant recommendations, quickly administering effective treatments, or performing profitable trades in fractions of a second. Such demands require algorithms to have the ability to learn in a dynamic environment, learn efficiently, and provide high quality solutions. Designing algorithms which meet such user demands poses significant challenges for researchers. In this thesis, we design and analyze machine learning algorithms which interact with the environment and specifically study two aspects which can help alleviate challenges: (1) learning online where the algorithm selects an action after which it receives the outcome (i.e., loss) of selecting such an action and (2) using the structure (sparsity, group sparsity, low-rankness) of a solution or user model. We explore such aspects under two feedback models: full and bandit information. With full information feedback, the algorithm observes the loss of each possible action it could have selected, for example, a trading algorithm can observe the price of each stock it could have invested in at the end of the day. With bandit information feedback, the algorithm can only observe the loss of the action taken, for example, a medical treatment algorithm can only observe whether a patient's health improved for the treatment provided. We measure the performance of our algorithms by their regret which is the difference between the cumulative loss received by the algorithm and the cumulative loss received by the best fixed or time-varying actions in hindsight. In the first half of this thesis, we focus on full information settings and study online learning algorithms for general resource allocation problems motivated, in part, by applications in algorithmic trading. The first two topics we explore are controlling the cost of updating an allocation, for example, one's stock portfolio, and learning to allocate a resource across groups of objects such as stock market sectors. In portfolio selection, making frequent trades may incur huge amounts of transaction costs and hurt one's bottom line. Moreover, groups of stocks, may perform similarly and investing in a few groups may lead to higher returns. We design and analyze two efficient algorithms and present new theoretical regret bounds. Further, we experimentally show the algorithms earn more wealth than existing algorithms even with transaction costs. The third and fourth topics we consider are two different ways to control suitable measures of risk associated with a resource allocation. The first approach is through diversification and the second is through the concept of hedging where, in the application of portfolio selection, a trader borrows shares from the bank and holds both long and short positions. We design and analyze two efficient online learning algorithms which either diversify across groups of stocks or hedge between individual stocks. We establish standard regret bounds and show experimentally our algorithms earn more wealth, in some cases orders of magnitude more, than existing algorithms and incur less risk. In the second half of this thesis, we focus on bandit information settings and how to use the structure of a user model to design algorithms with theoretically sharper regret bounds. We study the stochastic linear bandit problem which generalizes the widely studied multi-armed bandit. In the multi-armed bandit, an algorithm repeatedly selects an arm (action) from a finite decision set after which it receives a stochastic loss. In the stochastic linear bandit, arms are selected from a decision set with infinitely many arms (e.g., vectors from a compact set) and the loss is a stochastic linear function parameterized by an unknown vector. The first topic we explore is how the regret scales when the unknown parameter is structured (e.g., sparse, group sparse, low-rank). We design and analyze an algorithm which uses the structure to construct tight confidence sets which contain the unknown parameter with high-probability which leads to sharp regret bounds. The second topic we explore is how to generalize the previous algorithm to non-linear losses often used in Generalized Linear Models. We design and analyze a similar algorithm and show the regret is of the same order as with linear losses.Item Sustainability Policy Audit(Hubert H. Humphrey School of Public Affairs, 2012-05-18) Berrens, Christopher; Brown, Rebecca; Dirnberger, Amanda; Johnson, Nicholas