This thesis addresses the problem of named entity recognition (NER) in music-related search queries. NER is the task of identifying keywords in text and classifying them into predefined categories. Previous work in the field has mainly focused on longer documents of editorial texts. However, in recent years, the application of NER for queries has attracted increased attention. This task is, however, acknowledged to be challenging due to queries being short, ungrammatical and containing minimal linguistic context. The usage of NER for queries is especially useful for the implementation of natural language queries in domain-specific search applications. These applications are often backed by a database, where the query format otherwise is restricted to keyword search or the usage of a formal query language. In this thesis, two techniques for NER for music-related queries are evaluated; a conditional random field based solution and a probabilistic solution based on context words. As a baseline, the most elementary implementation of NER, commonly applied on editorial text, is used. Both of the evaluated approaches outperform the baseline and demonstrate an overall F1 score of 79.2% and 63.4% respectively. The experimental results show a high precision for the probabilistic approach and the conditional random field based solution demonstrates an F1 score comparable to previous studies from other domains. / Denna avhandling redogör för identifiering av namngivna enheter i musikrelaterade sökfrågor. Identifiering av namngivna enheter innebär att extrahera nyckelord från text och att klassificera dessa till någon av ett antal förbestämda kategorier. Tidigare forskning kring ämnet har framför allt fokuserat på längre redaktionella dokument. Däremot har intresset för tillämpningar på sökfrågor ökat de senaste åren. Detta anses vara ett svårt problem då sökfrågor i allmänhet är korta, grammatiskt inkorrekta och innehåller minimal språklig kontext. Identifiering av namngivna enheter är framför allt användbart för domänspecifika sökapplikationer där målet är att kunna tolka sökfrågor skrivna med naturligt språk. Dessa applikationer baseras ofta på en databas där formatet på sökfrågorna annars är begränsat till att enbart använda nyckelord eller användande av ett formellt frågespråk. I denna avhandling har två tekniker för identifiering av namngivna enheter för musikrelaterade sökfrågor undersökts; en metod baserad på villkorliga slumpfält (eng. conditional random field) och en probabilistisk metod baserad på kontextord. Som baslinje har den mest grundläggande implementationen, som vanligtvis används för redaktionella texter, valts. De båda utvärderade metoderna presterar bättre än baslinjen och ges ett F1-värde på 79,2% respektive 63,4%. De experimentella resultaten visar en hög precision för den probabilistiska implementationen och metoden ba- serad på villkorliga slumpfält visar på resultat på en nivå jämförbar med tidigare studier inom andra domäner.

Thesis (PhD)-- Stellenbosch University, 2013. / ENGLISH ABSTRACT: This dissertation considers automatic segmentation of the left cardiac ventricle in short axis magnetic resonance images. The presence of papillary muscles near the endocardium border makes simple threshold based segmentation difficult. The endo- and epicardium are modelled as two series of radii which are inter-related using features describing shape and motion. Image features are derived from edge information from human annotated images. The features are combined within a Conditional Random Field (CRF) – a discriminatively trained probabilistic model. Loopy belief propagation is used to infer segmentations when an unsegmented video sequence is given. Powell’s method is applied to find CRF parameters by minimising the difference between ground truth annotations and the inferred contours. We also describe how the endocardium centre points are calculated from a single human-provided centre point in the first frame, through minimisation of frame alignment error. We present and analyse the results of segmentation. The algorithm exhibits robustness against inclusion of the papillary muscles by integrating shape and motion information. Possible future improvements are identified. / AFRIKAANSE OPSOMMING: Hierdie proefskrif bespreek die outomatiese segmentasie van die linkerhartkamer in kortas snit magnetiese resonansie beelde. Die teenwoordigheid van die papillêre spiere naby die endokardium grens maak eenvoudige drumpel gebaseerde segmentering moeilik. Die endo- en epikardium word gemodelleer as twee reekse van die radiusse wat beperk word deur eienskappe wat vorm en beweging beskryf. Beeld eienskappe word afgelei van die rand inligting van mens-geannoteerde beelde. Die funksies word gekombineer binne ’n CRF (Conditional Random Field) – ’n diskriminatief afgerigte waarskynlikheidsverdeling. “Loopy belief propagation” word gebruik om segmentasies af te lei wanneer ’n ongesegmenteerde video verskaf word. Powell se metode word toegepas om CRF parameters te vind deur die minimering van die verskil tussen mens geannoteerde segmentasies en die afgeleide kontoere. Ons beskryf ook hoe die endokardium se middelpunte bereken word vanaf ’n enkele mens-verskafte middelpunt in die eerste raam, deur die minimering van ’n raambelyningsfout. Ons analiseer die resultate van segmentering. Die algoritme vertoon robuustheid teen die insluiting van die papillêre spiere deur die integrasie van inligting oor die vorm en die beweging. Moontlike toekomstige verbeterings word geïdentifiseer.

General Circulation Models (GCMs), which are mathematical models based on principles of fluid dynamics, thermodynamics and radiative transfer, are the most reliable tools available for projecting climate change. However, the spatial scale on which typical GCMs operate is very coarse as compared to that of a hydrologic process and hence, the output from a GCM cannot be directly used in hydrologic models. Statistical Downscaling (SD) derives a statistical or empirical relationship between the variables simulated by the GCM (predictors) and a point-scale meteorological series (predictand). In this work, a new downscaling model called CRF-downscaling model, is developed where the conditional distribution of the hydrologic predictand sequence, given atmospheric predictor variables, is represented as a conditional random field (CRF) to downscale the predictand in a probabilistic framework. Features defined in the downscaling model capture information about various factors influencing precipitation such as circulation patterns, temperature and pressure gradients and specific humidity levels. Uncertainty in prediction is addressed by projecting future cumulative distribution functions (CDFs) for a number of most likely precipitation sequences. Direct classification of dry/wet days as well as precipitation amount is achieved within a single modeling framework, and changes in the non-parametric distribution of precipitation and dry and wet spell lengths are projected. Application of the method is demonstrated with the case study of downscaling to daily precipitation in the Mahanadi basin in Orissa, with the A1B scenario of the MIROC3.2 GCM from the Center for Climate System Research (CCSR), Japan. An uncertainty modeling framework is presented in this work, which combines GCM, scenario and downscaling uncertainty using the Dempster-Shafer (D-S) evidence theory for representing and combining uncertainty. The methodology for combining uncertainties is applied to projections of hydrologic drought in terms of monsoon standardized streamflow index (SSFI-4) from streamflow projections for the Mahanadi river at Hirakud. The results from the work indicate an increasing probability of extreme, severe and moderate drought and decreasing probability of normal to wet conditions, as a result of a decrease in monsoon streamflow in the Mahanadi river due to climate change. In most studies to date, the nature of the downscaling relationship is assumed stationary, or remaining unchanged in a future climate. In this work, an uncertainty modeling framework is presented in which, in addition to GCM and scenario uncertainty, uncertainty in the downscaling relationship itself is explored by linking downscaling with changes in frequencies of modes of natural variability. Downscaling relationships are derived for each natural variability cluster and used for projections of hydrologic drought. Each projection is weighted with the future projected frequency of occurrence of that cluster, called ‘cluster-linking’, and scaled by the GCM performance with respect to the associated cluster for the present period, called ‘frequency scaling’. The uncertainty modeling framework is applied to a case study of projections of hydrologic drought or SSFI-4 classifications, using projected streamflows for the Mahanadi river at Hirakud. It is shown that a stationary downscaling relationship will either over- or under-predict downscaled hydrologic variable values and associated uncertainty. Results from the work show improved agreement between GCM predictions at the regional scale, which are validated for the 20th century, implying that frequency scaling and cluster-linking may indeed be a valid method for constraining uncertainty. To assess the impact of climate change on reservoir performance, in this study, a range of integrated hydrologic scenarios are projected for the future. The hydrologic scenarios incorporate increased irrigation demands; rule curves dictated by increased need for flood storage and downscaled projections of streamflow from an ensemble of GCMs and emission scenarios. The impact of climate change on multipurpose reservoir performance is quantified, using annual hydropower and RRV criteria, under GCM and scenario uncertainty. The ‘business-as-usual’ case using Standard Operating Policy (SOP) is studied initially for quantifying impacts. Adaptive Stochastic Dynamic Programming (SDP) policies are subsequently derived for the range of future hydrologic scenarios, with the objective of maximizing reliabilities with respect to multiple reservoir purposes of hydropower, irrigation and flood control. It is shown that the hydrologic impact of climate change is likely to result in decreases in performance criteria and annual hydropower generation for Hirakud reservoir. Adaptive policies show that a marginal reduction in irrigation and flood control reliability can achieve increased hydropower reliability in future. Hence, reservoir rules for flood control may have to be revised in the future.