Voice user interfaces have been growing in popularity and with them an interest for open vocabulary keyword spotting. In this thesis we focus on one particular approach to open vocabulary keyword spotting, query by example keyword spotting. Three types of query by example keyword spotting approaches are described and evaluated: sequence distances, speech to phonemes and deep distance learning. Evaluation is done on a series of custom tasks designed to measure a variety of aspects. The Google Speech Commands benchmark is used for evaluation as well, this to make it more comparable to existing works. From the results, the deep distance learning approach seem most promising in most environments except when memory is very constrained; in which sequence distances might be considered. The speech to phonemes methods is lacking in the usability evaluation. / Röstgränssnitt har växt i populäritet och med dem ett intresse för öppenvokabulärnyckelordsigenkänning. I den här uppsatsen fokuserar vi på en specifik form av öppenvokabulärnyckelordsigenkänning, den s.k nyckelordsigenkänning- genom- exempel. Tre typer av nyckelordsigenkänning- genom- exempel metoder beskrivs och utvärderas: sekvensavstånd, tal till fonem samt djupavståndsinlärning. Utvärdering görs på konstruerade uppgifter designade att mäta en mängd olika aspekter hos metoderna. Google Speech Commands data används för utvärderingen också, detta för att göra det mer jämförbart mot existerade arbeten. Från resultaten framgår det att djupavståndsinlärning verkar mest lovande förutom i miljöer där resurser är väldigt begränsade; i dessa kan sekvensavstånd vara av intresse. Tal till fonem metoderna visar brister i användningsuvärderingen.

Relations that hold between causes and their effects are fundamental for a wide range of different sectors. Automatically finding sentences that express such relations may for example be of great interest for the economy or political institutions. However, for many languages other than English, a lack of training resources for this task needs to be dealt with. In recent years, large, pretrained transformer-based model architectures have proven to be very effective for tasks involving cross-lingual transfer such as cross-lingual language inference, as well as multilingual named entity recognition, POS-tagging and dependency parsing, which may hint at similar potentials for causality detection. In this thesis, we define causality detection as a binary labelling problem and use cross-lingual transfer to alleviate data scarcity for German and Swedish by using three different classifiers that make either use of multilingual sentence embeddings obtained from a pretrained encoder or pretrained multilingual language models. The source languages in most of our experiments will be English, for Swedish we however also use a small German training set and a combination of English and German training data.  We try out zero-shot transfer as well as making use of limited amounts of target language data either as a development set or as additional training data in a few-shot setting. In the latter scenario, we explore the impact of varying sizes of training data. Moreover, the problem of data scarcity in our situation also makes it necessary to work with data from different annotation projects. We also explore how much this would impact our result. For German as a target language, our results in a zero-shot scenario expectedly fall short in comparison with monolingual experiments, but F1-macro scores between 60 and 65 in cases where annotation did not differ drastically still signal that it was possible to transfer at least some knowledge. When introducing only small amounts of target language data, already notable improvements were observed and with the full German training data of about 3,000 sentences combined with the most suitable English data set, the performance for German in some scenarios even almost matches the state of the art for monolingual experiments on English. The best zero-shot performance on the Swedish data was even outperforming the scores achieved for German. However, due to problems with the additional Swedish training data, we were not able to improve upon the zero-shot performance in a few-shot setting in a similar manner as it was the case for German.

We utilize the exact factorization of the electron-nuclear wave function [Abedi et al., PRL 105 123002 (2010)] to illuminate several aspects of laser-driven molecular dynamics in intense femtosecond laser pulses. Above factorization allows for a splitting of the full molecular wave function and leads to a time-dependent Schrödinger equation for the nuclear subsystem alone which is exact in the sense that the absolute square of the corresponding, purely nuclear, wave function yields the exact nuclear N-body density of the full electron-nuclear system. As one remarkable feature, this factorization provides the exact classical force, the force which contains the highest amount of electron-nuclear correlations that can be retained in the quantum-classical limit of the electron-nuclear system. We re-evaluate the classical limit of the nuclear Schrödinger equation from the perspective of the exact factorization, and address the long-standing question of the validity of the popular quantum-classical surface hopping approach in laserdriven cases. In particular, our access to the exact classical force allows for an elaborate evaluation of the various and completely different potential energy surfaces frequently applied in surface hopping calculations. The highlight of this work consists in a generalization of the exact factorization and its application to the laser-driven molecular wave function in the Floquet picture, where the molecule and the laser form an united quantum system exhibiting its own Hilbert space. This particular factorization enables us to establish an analytic connection between the exact nuclear force and Floquet potential energy surfaces. Complementing above topics, we combine different well-known and proven methods to give a systematic study of molecular dissociation mechanisms for the complicated electric fields provided by modern attosecond laser technology.:Contents Introduction 1 The exact factorization of time-dependent wave functions 1.1 Concern and state of the art 1.2 The exact factorization of the electron-nuclear wave function 1.3 The generalized exact factorization 1.4 The exact factorization for coupled harmonic oscillators 1.5 The exact factorization for a single particle with spin 1.6 The exact factorization of the laser-driven electron-nuclear wave function in the Floquet picture 1.7 Summary and conclusion 2 Quantum-classical molecular dynamics from an exact factorization perspective 2.1 Concern and state of the art 2.2 The exact nuclear TDSE 2.3 The Wigner-Moyal equation for the nuclear TDSE and its classical limit 2.4 The Bohmian formulation of the nuclear TDSE and its classical limit 2.5 Comparative calculations 2.5.1 Scenario 1: stationary states 2.5.2 Scenario 2: laser-driven dynamics 2.6 Summary and conclusion 3 Surface hopping in laser-driven molecular dynamics 3.1 Concern and state of the art 3.2 Surface hopping 3.3 Quantum-classical dynamics on the EPES 3.4 The benchmark model and its potential energy surfaces 3.5 Surface hopping in laser-driven molecular dynamics 3.6 Summary and conclusion 4 Beyond the limit of the Floquet picture: molecular dissociation in few-cycle laser pulses 4.1 Concern and state of the art 4.2 Theoretical few-cycle pulses 4.3 Calculation of dissociation probabilities 4.4 Dissociation in few-cycle pulses 4.4.1 Dissociation in half-cycle pulses 4.4.2 Dissociation in few-cycle pulses 4.5 Dissociation in realistic attosecond pulses 4.6 Summary and conclusion Outlook Appendices A List of abbreviations B Numerical details C Calculating electronic observables within quantum-classical molecular dynamics D Ionization in few-cycle pulses E Modeling an optical attosecond pulse Bibliography

In forensic casework analysis it is necessary to obtain genetic profiles from increasingly smaller amounts of biological material left behind by perpetrators of crime. The ability to obtain profiles from trace biological evidence is demonstrated with so-called ‘touch DNA evidence’ which is perceived to be the result of DNA obtained from shed skin cells transferred from donor to an object or person during physical contact. However, the current method of recovery of trace DNA involves cotton swabs or adhesive tape to sample an area of interest. This "blindswabbing" approach may result in the recovery of biological material from different individuals resulting in admixed DNA profiles which are often difficult to interpret. Profiles recovered from these samples are reported to be from shed skin cells with no biological basis for that determination. A specialized approach for the isolation of single or few cells from ‘touch DNA evidence’ is necessary to improve the analysis and interpretation of recovered profiles. Here we describe the development of optimized and robust micro volume PCR reactions (1-5 μL) to improve the sensitivity and efficiency of ‘touch DNA’ analysis. These methods will permit not only the recovery of the genetic profile of the donor of the biological material, but permit an identification of the tissue source of origin using mRNA profiling. Results showed that the 3.5 uL amplification volume, a fraction of the standard 25 uL amplification volume, was the most ideal volume for the DNA assay, as it had very minimal evaporation with a 50% profile recovery rate at a single cell equivalent input (~5 pg) with reducing amplification volume alone. Findings for RNA showed that by reducing both amplification steps, reverse transcriptase PCR (20 uL) and body fluid multiplex PCR (25 uL), to iv 5 uL, ideal results were obtained with an increase in sensitivity and detection of six different body fluids down to 50 pg. Once optimized at the trace level, the assays were applied to the collection of single and few cells. DNA findings showed that about 40% of a full profile could be recovered from a single buccal cell, with nearly 80% of a full profile recovered from only two cells. RNA findings from collected skin particles of "touched" surfaces showed accurate skin detection down to 25 particles and detection in one clump of particles. The profiles recovered were of high quality and similar results were able to be replicated through subsequent experiments. More studies are currently underway to optimize these developed assays to increase profile recovery at the single cell level. Methods of doing so include comparing different locations on touched surfaces for highest bio-particle recovery and the development of physical characteristics of bio-particles that would provide the most ideal results

Address parsing is the process of splitting an address string into its different address components, such as street name, street number, et cetera. Address parsing has been quite extensively researched and there exist some state-ofthe-art address parsing solutions, mostly unilingual. In more recent years research has emerged which focuses on multinational address parsing and deep architecture address parsers have been used to achieve state-of-the-art performance on multinational address data. However, training these deep architectures for address parsing requires a rather large amount of address data which is not always accessible. Generally within Natural Language Processing (NLP) data is difficult to come by and most of the NLP data available consists of data from about only 20 of the approximately 7000 languages spoken around the world, so-called high-resource languages. This also applies to address data, which can be difficult to come by for some of the so-called low-resource languages of the world for which little or no NLP data exists. To attempt to deal with the lack of address data availability for some of the less spoken languages of the world, the current project investigates the potential of FewShot Learning (FSL) for multinational address parsing. To investigate this, two few-shot transfer learning models are implemented, both implementations consist of a fine-tuned pre-trained language model (PTLM). The difference between the two models is the PTLM used, which were the multilingual language models mBERT and XLM-R, respectively. The two PTLMs are finetuned using a linear classifier layer to then be used as multinational address parsers. The two models are trained and their results are compared with a state-of-the-art multinational address parser, Deepparse, as well as with each other. Results show that the two models do not outperform Deepparse, but they do show promising results, not too far from what Deepparse achieves on holdout and zero-shot datasets. On a mix of low- and high-resource language address data, both models perform well and achieve over 96% on the overall F1-score. Out of the two models used for implementation, XLM-R achieves significantly better results than mBERT and can therefore be considered the more appropriate PTLM to use for multinational FSL address parsing. Based on these results the conclusion is that there is great potential for FSL within the field of multinational address parsing and that general FSL methods can be used and perform well on multinational address parsing tasks. / Adressavkodning är processen att dela upp en adresssträng i dess olika adresskomponenter såsom gatunamn, gatunummer, et cetera. Adressavkodning har undersökts ganska omfattande och det finns några toppmoderna adressavkodningslösningar, mestadels enspråkiga. Senaste åren har forskning fokuserad på multinationell adressavkodning börjat dyka upp och djupa arkitekturer för adressavkodning har använts för att uppnå toppmodern prestation på multinationell adressdata. Att träna dessa arkitekturer kräver dock en ganska stor mängd adressdata, vilket inte alltid är tillgängligt. Det är generellt svårt att få tag på data inom naturlig språkbehandling och majoriteten av den data som är tillgänglig består av data från endast 20 av de cirka 7000 språk som används runt om i världen, så kallade högresursspråk. Detta gäller även för adressdata, vilket kan vara svårt att få tag på för vissa av världens så kallade resurssnåla språk för vilka det finns lite eller ingen data för naturlig språkbehandling. För att försöka behandla denna brist på adressdata för några av världens mindre talade språk undersöker detta projekt om det finns någon potential för inlärning med få exempel för multinationell adressavkodning. För detta implementeras två modeller för överföringsinlärning med få exempel genom finjustering av förtränade språkmodeller. Skillnaden mellan de två modellerna är den förtränade språkmodellen som används, mBERT respektive XLM-R. Båda modellerna finjusteras med hjälp av ett linjärt klassificeringsskikt för att sedan användas som multinationella addressavkodare. De två modellerna tränas och deras resultat jämförs med en toppmodern multinationell adressavkodare, Deepparse. Resultaten visar att de två modellerna presterar båda sämre än Deepparse modellen, men de visar ändå lovande resultat, inte långt ifrån vad Deepparse uppnår för både holdout och zero-shot dataset. Vidare, så presterar båda modeller bra på en blandning av adressdata från låg- och högresursspråk och båda modeller uppnår över 96% övergripande F1-score. Av de två modellerna uppnår XLM-R betydligt bättre resultat än mBERT och kan därför anses vara en mer lämplig förtränad språkmodell att använda för multinationell inlärning med få exempel för addressavkodning. Utifrån dessa resultat dras slutsatsen att det finns stor potential för inlärning med få exempel inom området multinationall adressavkodning, samt att generella metoder för inlärning med få exempel kan användas och preseterar bra på multinationella adressavkodningsuppgifter.

No description available.

No description available.

To label words of interest into a predefined set of named entities have traditionally required a large amount of labeled in-domain data. Recently, the availability of pre-trained transformer-based language models have enabled multiple natural language processing problems to utilize transfer learning techniques to construct machine learning models with less task-specific labeled data. In this thesis, the impact of data augmentation when training a pre-trained transformer-based model to adapt to a named entity recognition task with few labeled sentences is explored. The experimental results indicate that data augmentation increases performance of the trained models, however the data augmentation is shown to have less impact when more labeled data is available. In conclusion, data augmentation has been shown to improve performance of pre-trained named entity recognition models when few labeled sentences are available for training. / Att kategorisera ord som tillhör någon av en mängd förangivna entiteter har traditionellt krävt stora mängder förkategoriserad områdesspecifik data. På senare år har det tillgängliggjorts förtränade språkmodeller som möjliggjort för språkprocesseringsproblem att lösas med en mindre mängd områdesspecifik kategoriserad data. I den här uppsatsen utforskas datautöknings påverkan på en maskininlärningsmodell för identifiering av namngivna entiteter. De experimentella resultaten indikerar att datautökning förbättrar modellerna, men att inverkan blir mindre när mer kategoriserad data är tillgänglig. Sammanfattningsvis så kan datautökning förbättra modeller för identifiering av namngivna entiteter när få förkategoriserade meningar finns tillgängliga för träning.

This thesis investigates the possibility of automating the classification of post-feedback mathematical reasoning styles, Creative Mathematical Reasoning (CMR) and Algorithmic Reasoning (AR), using prompt-based classification with a Large Language Model (LLM). The study, conducted in collaboration with the Department of Science and Mathematics Education of Umeå University, aims to enhance the efficiency of pedagogical research by reducing the manual labor involved in classifying student responses. The thesis utilizes a dataset of 40 expert-labeled student mathematical solutions, incorporating feedback interactions to assess shifts in reasoning post-feedback. Various prompting methods, including definitions-only and examples-inclusive prompts, were systematically tested to determine their effectiveness in classifying reasoning styles. The classification performance was measured using accuracy, F1-score, and Cohen’s kappa. Results indicate that definitionbased prompts performed robustly, achieving moderate to strong inter-rater agreement. The study also explored the impact of output formats and found that allowing the LLM to classify uncertain cases as indeterminate could potentially automate about 25% of the classification tasks without compromising performance. This thesis underscores the potential of LLMs in automating complex cognitive task classifications in educational research, suggesting further exploration into optimal prompting strategies and reliability enhancements for practical applications. / Denna uppsats undersöker möjligheten att automatisera klassificeringen av matematiska resonemangstyper efter feedback, Kreativt Matematiskt Resonemang (CMR) och Algoritmiskt Resonemang (AR), med hjälp av promptbaserad klassificering med en stor språkmodell (LLM). Studien, som genomfördes i samarbete med Institutionen för naturvetenskapernas och matematikens didaktik vid Umeå universitet, syftar till att öka effektiviteten i pedagogisk forskning genom att minska det manuella arbetet som krävs för att klassificera studenters matematiska resonemang. Uppsatsen använder ett dataset med 40 matematiska lösningar från studenter, klassificerade av experter. Dessa lösningar inkluderar feedback-interaktioner för att bedöma förändringar i resonemang efter feedback. Olika promptmetoder, innehållandes enbart definitioner och exempel-inkluderande promptar, testades systematiskt för att avgöra deras effektivitet vid klassificering av resonemangsstilar. Klassificeringsprestandan mättes med hjälp av accuracy, F1-score och Cohen’s kappa. Resultaten visar att promptar baserade på definitioner hade en robust prestanda och uppnådde måttlig till stark överensstämmelse mellan bedömare. Studien undersökte också påverkan av utdataformat och fann att genom att tillåta LLM att klassificera osäkra fall som obestämdbarkunde cirka 25% av klassificeringsuppgifterna automatiseras utan att kompromissa med prestandan. Denna avhandling framhäver potentialen hos LLMs att automatisera komplexa kognitiva uppgiftsklassificeringar inom utbildningsforskning och föreslår vidare studier av optimala promptstrategier och tillförlitlighetsförbättringar för praktiska tillämpningar.

No / Here we report a hierarchical composite structure composed of few-layers molybdenum disulfide nanosheets supported by vertical graphene on conductive carbon cloth (MDNS/VG/CC) for high-performance electrochemical hydrogen evolution reaction (HER). In the fabrication, 3D vertical graphene is first prepared on carbon cloth by a micro-wave plasma enhanced chemical vapor deposition (MPCVD) and then few-layers MoS2 nanosheets are in-situ synthesized on the surface of the vertical graphene through a simple hydrothermal reaction. This integrated catalyst exhibits an excellent HER electrocatalytic activity including an onset potential of 50 mV, an overpotential at 10 mA cm(-2) (eta(10)) of 78 mV, a Tafel slop of 53 mV dec(-1), and an excellent cycling stability in acid solution. The excellent catalytic performance can be ascribed to the abundant active edges provided by the vertical MoS2 nanosheets, as well as the effective electron transport route provided by the graphene arrays on the conductive substrate. Moreover, the vertical graphene offers robust anchor sites for MoS2 nanosheets and appropriate intervals for electrolyte infiltration. This not only benefits hydrogen convection and release but also avoids the damaging or restacking of catalyst in electrochemical processes. / This work was financially supported by the National Natural Science Foundation of China (Grant nos. 61176007, 51372213, and 51402343).