39 resultados para Bayesian approaches
Resumo:
Accelerator mass spectrometry (AMS) is an ultrasensitive technique for measuring the concentration of a single isotope. The electric and magnetic fields of an electrostatic accelerator system are used to filter out other isotopes from the ion beam. The high velocity means that molecules can be destroyed and removed from the measurement background. As a result, concentrations down to one atom in 10^16 atoms are measurable. This thesis describes the construction of the new AMS system in the Accelerator Laboratory of the University of Helsinki. The system is described in detail along with the relevant ion optics. System performance and some of the 14C measurements done with the system are described. In a second part of the thesis, a novel statistical model for the analysis of AMS data is presented. Bayesian methods are used in order to make the best use of the available information. In the new model, instrumental drift is modelled with a continuous first-order autoregressive process. This enables rigorous normalization to standards measured at different times. The Poisson statistical nature of a 14C measurement is also taken into account properly, so that uncertainty estimates are much more stable. It is shown that, overall, the new model improves both the accuracy and the precision of AMS measurements. In particular, the results can be improved for samples with very low 14C concentrations or measured only a few times.
Resumo:
Nucleation is the first step of a first order phase transition. A new phase is always sprung up in nucleation phenomena. The two main categories of nucleation are homogeneous nucleation, where the new phase is formed in a uniform substance, and heterogeneous nucleation, when nucleation occurs on a pre-existing surface. In this thesis the main attention is paid on heterogeneous nucleation. This thesis wields the nucleation phenomena from two theoretical perspectives: the classical nucleation theory and the statistical mechanical approach. The formulation of the classical nucleation theory relies on equilibrium thermodynamics and use of macroscopically determined quantities to describe the properties of small nuclei, sometimes consisting of just a few molecules. The statistical mechanical approach is based on interactions between single molecules, and does not bear the same assumptions as the classical theory. This work gathers up the present theoretical knowledge of heterogeneous nucleation and utilizes it in computational model studies. A new exact molecular approach on heterogeneous nucleation was introduced and tested by Monte Carlo simulations. The results obtained from the molecular simulations were interpreted by means of the concepts of the classical nucleation theory. Numerical calculations were carried out for a variety of substances nucleating on different substances. The classical theory of heterogeneous nucleation was employed in calculations of one-component nucleation of water on newsprint paper, Teflon and cellulose film, and binary nucleation of water-n-propanol and water-sulphuric acid mixtures on silver nanoparticles. The results were compared with experimental results. The molecular simulation studies involved homogeneous nucleation of argon and heterogeneous nucleation of argon on a planar platinum surface. It was found out that the use of a microscopical contact angle as a fitting parameter in calculations based on the classical theory of heterogeneous nucleation leads to a fair agreement between the theoretical predictions and experimental results. In the presented cases the microscopical angle was found to be always smaller than the contact angle obtained from macroscopical measurements. Furthermore, molecular Monte Carlo simulations revealed that the concept of the geometrical contact parameter in heterogeneous nucleation calculations can work surprisingly well even for very small clusters.
Resumo:
This dissertation empirically explores the relations among three theoretical perspectives: university students approaches to learning, self-regulated learning, as well as cognitive and attributional strategies. The relations were quantitatively studied from both variable- and person-centered perspectives. In addition, the meaning that students gave to their disciplinary choices was examined. The general research questions of the study were: 1) What kinds of relationships exist among approaches to learning, regulation of learning, and cognitive and attributional strategies? What kinds of cognitive-motivational profiles can be identified among university students, and how are such profiles related to study success and well-being? 3) How do university students explain their disciplinary choices? Four empirical studies addressed these questions. Studies I, II, and III were quantitative, applying self-report questionnaires, and Study IV was qualitative in nature. Study I explored relations among cognitive strategies, approaches to learning, regulation of learning, and study success by using correlations and a K-means cluster analysis. The participants were 366 students from various faculties at different phases of their studies. The results showed that all the measured constructs were logically related to each other in both variable- and person-centered approaches. Study II further examined what kinds of cognitive-motivational profiles could be identified among first-year university students (n=436) in arts, law, and agriculture and forestry. Differences in terms of study success, exhaustion, and stress among students with differing profiles were also looked at. By using a latent class cluster analysis (LCCA), three groups of students were identified: non-academic (34%), self-directed (35%), and helpless students (31%). Helpless students reported the highest levels of stress and exhaustion. Self-directed students received the highest grades. In Study III, cognitive-motivational profiles were identified among novice teacher students (n=213) using LCCA. Well-being, epistemological beliefs, and study success were looked at in relation to the profiles. Three groups of students were found: non-regulating (50%), self-directed (35%), and non-reflective (22%). Self-directed students again received the best grades. Non-regulating students reported the highest levels of stress and exhaustion, the lowest level of interest, and showed the strongest preference for certain and practical knowledge. Study IV, which was qualitative in nature, explored how first-year students (n = 536 ) in three fields of studies, arts, law, and veterinary medicine explained their disciplinary choices. Content analyses showed that interest appeared to be a common concept in students description of their choices across the three faculties. However, the objects of interest of the freshmen appeared rather unspecified. Veterinary medicine and law students most often referred to future work or a profession, whereas only one-fifth of the arts students did so. The dissertation showed that combining different theoretical perspectives and methodologies enabled us to build a rich picture of university students cognitive and motivational predispositions towards studying and learning. Further, cognitive-emotional aspects played a significant role in studying, not only in relation to study success, but also in terms of well-being. Keywords: approaches to learning, self-regulation, cognitive and attributional strategies, university students
Resumo:
The Earth s climate is a highly dynamic and complex system in which atmospheric aerosols have been increasingly recognized to play a key role. Aerosol particles affect the climate through a multitude of processes, directly by absorbing and reflecting radiation and indirectly by changing the properties of clouds. Because of the complexity, quantification of the effects of aerosols continues to be a highly uncertain science. Better understanding of the effects of aerosols requires more information on aerosol chemistry. Before the determination of aerosol chemical composition by the various available analytical techniques, aerosol particles must be reliably sampled and prepared. Indeed, sampling is one of the most challenging steps in aerosol studies, since all available sampling techniques harbor drawbacks. In this study, novel methodologies were developed for sampling and determination of the chemical composition of atmospheric aerosols. In the particle-into-liquid sampler (PILS), aerosol particles grow in saturated water vapor with further impaction and dissolution in liquid water. Once in water, the aerosol sample can then be transported and analyzed by various off-line or on-line techniques. In this study, PILS was modified and the sampling procedure was optimized to obtain less altered aerosol samples with good time resolution. A combination of denuders with different coatings was tested to adsorb gas phase compounds before PILS. Mixtures of water with alcohols were introduced to increase the solubility of aerosols. Minimum sampling time required was determined by collecting samples off-line every hour and proceeding with liquid-liquid extraction (LLE) and analysis by gas chromatography-mass spectrometry (GC-MS). The laboriousness of LLE followed by GC-MS analysis next prompted an evaluation of solid-phase extraction (SPE) for the extraction of aldehydes and acids in aerosol samples. These two compound groups are thought to be key for aerosol growth. Octadecylsilica, hydrophilic-lipophilic balance (HLB), and mixed phase anion exchange (MAX) were tested as extraction materials. MAX proved to be efficient for acids, but no tested material offered sufficient adsorption for aldehydes. Thus, PILS samples were extracted only with MAX to guarantee good results for organic acids determined by liquid chromatography-mass spectrometry (HPLC-MS). On-line coupling of SPE with HPLC-MS is relatively easy, and here on-line coupling of PILS with HPLC-MS through the SPE trap produced some interesting data on relevant acids in atmospheric aerosol samples. A completely different approach to aerosol sampling, namely, differential mobility analyzer (DMA)-assisted filter sampling, was employed in this study to provide information about the size dependent chemical composition of aerosols and understanding of the processes driving aerosol growth from nano-size clusters to climatically relevant particles (>40 nm). The DMA was set to sample particles with diameters of 50, 40, and 30 nm and aerosols were collected on teflon or quartz fiber filters. To clarify the gas-phase contribution, zero gas-phase samples were collected by switching off the DMA every other 15 minutes. Gas-phase compounds were adsorbed equally well on both types of filter, and were found to contribute significantly to the total compound mass. Gas-phase adsorption is especially significant during the collection of nanometer-size aerosols and needs always to be taken into account. Other aims of this study were to determine the oxidation products of β-caryophyllene (the major sesquiterpene in boreal forest) in aerosol particles. Since reference compounds are needed for verification of the accuracy of analytical measurements, three oxidation products of β-caryophyllene were synthesized: β-caryophyllene aldehyde, β-nocaryophyllene aldehyde, and β-caryophyllinic acid. All three were identified for the first time in ambient aerosol samples, at relatively high concentrations, and their contribution to the aerosol mass (and probably growth) was concluded to be significant. Methodological and instrumental developments presented in this work enable fuller understanding of the processes behind biogenic aerosol formation and provide new tools for more precise determination of biosphere-atmosphere interactions.
Resumo:
Inelastic x-ray scattering spectroscopy is a versatile experimental technique for probing the electronic structure of materials. It provides a wealth of information on the sample's atomic-scale structure, but extracting this information from the experimental data can be challenging because there is no direct relation between the structure and the measured spectrum. Theoretical calculations can bridge this gap by explaining the structural origins of the spectral features. Reliable methods for modeling inelastic x-ray scattering require accurate electronic structure calculations. This work presents the development and implementation of new schemes for modeling the inelastic scattering of x-rays from non-periodic systems. The methods are based on density functional theory and are applicable for a wide variety of molecular materials. Applications are presented in this work for amorphous silicon monoxide and several gas phase systems. Valuable new information on their structure and properties could be extracted with the combination of experimental and computational methods.
Resumo:
We propose an efficient and parameter-free scoring criterion, the factorized conditional log-likelihood (ˆfCLL), for learning Bayesian network classifiers. The proposed score is an approximation of the conditional log-likelihood criterion. The approximation is devised in order to guarantee decomposability over the network structure, as well as efficient estimation of the optimal parameters, achieving the same time and space complexity as the traditional log-likelihood scoring criterion. The resulting criterion has an information-theoretic interpretation based on interaction information, which exhibits its discriminative nature. To evaluate the performance of the proposed criterion, we present an empirical comparison with state-of-the-art classifiers. Results on a large suite of benchmark data sets from the UCI repository show that ˆfCLL-trained classifiers achieve at least as good accuracy as the best compared classifiers, using significantly less computational resources.
Resumo:
Bayesian networks are compact, flexible, and interpretable representations of a joint distribution. When the network structure is unknown but there are observational data at hand, one can try to learn the network structure. This is called structure discovery. This thesis contributes to two areas of structure discovery in Bayesian networks: space--time tradeoffs and learning ancestor relations. The fastest exact algorithms for structure discovery in Bayesian networks are based on dynamic programming and use excessive amounts of space. Motivated by the space usage, several schemes for trading space against time are presented. These schemes are presented in a general setting for a class of computational problems called permutation problems; structure discovery in Bayesian networks is seen as a challenging variant of the permutation problems. The main contribution in the area of the space--time tradeoffs is the partial order approach, in which the standard dynamic programming algorithm is extended to run over partial orders. In particular, a certain family of partial orders called parallel bucket orders is considered. A partial order scheme that provably yields an optimal space--time tradeoff within parallel bucket orders is presented. Also practical issues concerning parallel bucket orders are discussed. Learning ancestor relations, that is, directed paths between nodes, is motivated by the need for robust summaries of the network structures when there are unobserved nodes at work. Ancestor relations are nonmodular features and hence learning them is more difficult than modular features. A dynamic programming algorithm is presented for computing posterior probabilities of ancestor relations exactly. Empirical tests suggest that ancestor relations can be learned from observational data almost as accurately as arcs even in the presence of unobserved nodes.