The sum-product network (SPN) is a recently-proposed generative, probabilistic model that is guaranteed to compute any joint or any marginal probability in time linear in the size of the model. An SPN is represented as a directed, acyclic graph (DAG) of sum and product nodes, with univariate probability distributions at the leaves. It is important to learn the structure of this DAG since the set of distributions representable by an SPN is constrained by it. We present the first batch structure learning algorithm for SPNs and show its advantage over learning the parameters of an SPN with fixed architecture. We propose a search algorithm for learning the structure of an SPN and show that its ability to learn a DAG-structured SPN makes it better for some tasks than algorithms that only learn tree-structured SPNs. We adapt the structure search algorithm to learn the structure of an SPN in the online setting and show that two other methods for online SPN structure learning are slower or learn models with lower likelihood. We also propose to combine SPNs with an autoencoder to model image data; this application of SPN structure learning shows that both models benefit from being combined.We are also the first to propose a distinction between nonmonotone and monotone SPNs, or SPNs with negative edge-weights and those without, respectively. We prove several important properties of nonmonotone SPNs, propose algorithms for learning a special class of nonmonotone SPN called the twin SPNs, and show that allowing negative edge-weights can help twin SPNs model some distributions more compactly than monotone SPNs.

This thesis discusses Kaplan, a free conformer searching package, available at github.com/PeaWagon/Kaplan / Conformer searching algorithms find minima in the Potential Energy Surface (PES) of a molecule, usually by following a torsion-driven approach. The minima represent conformers, which are interchangeable via free rotation around bonds. Conformers can be used as input to computational analyses, such as drug design, that can convey molecular reactivity, structure, and function. With an increasing number of rotatable bonds, finding optima in the PES becomes more complicated, as the dimensionality explodes. Kaplan is a new, free and open-source software package written by the author that uses a ring-based Evolutionary Algorithm (EA) to find conformers. The ring, which contains population members (or pmems), is designed to allow initial PES exploration, followed by exploitation of individual energy wells, such that the most energetically-favourable structures are returned. The strengths and weaknesses of existing publicly available conformer searchers are discussed, including Balloon, RDKit, Openbabel, Confab, Frog2, and Kaplan. Since RDKit is usually considered to be the best free package for conformer searching, its conformers for the amino acids were optimised using the MMFF94 forcefield and compared to the conformers generated by Kaplan. Amino acid conformers are well characterised, and provide insight for protein substructure. Of the 20 molecules, Kaplan found a lower energy minima for 12 of the structures and tied for 5 of them. Kaplan allows the user to specify which dihedrals (by atom indices) to optimise and angles to use, a feature that is not offered by other programs. The results from Kaplan were compared to a known dataset of amino acid conformers. Kaplan identified all 57 conformers of methionine to within 1.2Å, and found identical conformers for the 5 lowest-energy structures (i.e. within 0.083Å), following forcefield optimisation. / Thesis / Master of Science (MSc) / A conformer search affords the low-energy arrangements of atoms that can be obtained via rotation around bonds. Conformers provide insight about the chemical reactivity and physical properties of a molecule. With increasing molecule size, the number of possible conformers increases exponentially. To search the space of possible conformers, this thesis presents Kaplan, which is a software package that implements a novel directed, stochastic, sampling technique based on an Evolutionary Algorithm (EA). Kaplan uses a special type of EA that stores sets of conformers in a ring-based structure. Unlike other conformer-specific packages, Kaplan provides the means to analyse and interact with found conformers. Known conformers of amino acids are used to verify Kaplan. Other tools for generating conformers are discussed, including a comparison of freely available software. Kaplan effectively finds the conformers of small molecules, but requires additional parametrisation to find the conformers of mid-sized molecules, such as Penta-Alanine.

In this thesis, density functional theory (DFT) calculations and DFT based ab initio molecular dynamics simulations have been employed in order to gain insights into materials properties like diffusion, adsorption, catalysis, and structure. In transition metals, absorbed hydrogen atoms self-trap due to localization of metal d-electrons. The self-trapping state is shown to highly influence hydrogen diffusion in the classical over-barrier jump temperature region. Li diffusion in Li-N-H systems is investigated. The diffusion in Li3N is shown to be controlled by the concentration of vacancies. Exchanging one Li for H (Li2NH), gives a system where the diffusion no longer is dependent on the concentrations of vacancies, but instead on N-H rotations. Furthermore, exchanging another Li for H (LiNH2), results in a blockade of Li diffusion. For high-surface area hydrogen storage materials, metal organic frameworks and covalent organic frameworks, the hydrogen adsorption is studied. In metal organic frameworks, a Li-decoration is also suggested as a way to increase the hydrogen adsorption energy. In NaAlH4 doped with transition metals (TM), the hypothesis of TM-Al intermetallic alloys as the main catalytic species is supported. The source of the catalytic effect of carbon nanostructures on hydrogen desorption from NaAlH4 is shown to be the high electronegativity of the carbon nanostructures. A space-group optimized ab initio random structure search method is used to find a new ground state structure for BeC2 and MgC2. The fast change between the amorphous and the crystalline phase of GeSbTe phase-change materials is suggested to be due to the close resemblance between the local amorphous structure and the crystalline structure. Finally, we show that more than 80% of the voltage in the lead acid battery is due to relativistic effects. / Felaktigt tryckt som Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Science and Technology 702