The rate of Emergency Department (ED) visits due to mental health and drug abuse among children and youth has been increasing for more than a decade and is projected to become the leading cause of ED visits. Identifying high-risk patients well before an ED visit will enable mental health care providers to better predict ED resource utilization, improve their service, and ultimately reduce the risk of a future ED visit. Many studies in the literature utilized medical history to predict future hospitalization. However, in mental health care, the medical history of new patients is not always available from the first visit and it is crucial to identify high risk patients from the beginning as the rate of drop-out is very high in mental health treatment. In this study, a new approach of creating a text representation of questionnaire data for deep learning analysis is proposed. Employing this new text representation has enabled us to use transfer learning and develop a deep Natural Language Processing (NLP) model that estimates the possibility of 6-month ED visit among children and youth using mental health patient reported outcome measures (PROM). The proposed method achieved an Area Under Receiver Operating Characteristic Curve of 0.75 for classification of 6-month ED visit. In addition, a novel method was proposed to identify the words that carry the highest amount of information related to the outcome of the deep NLP models. This measurement of word information using Entropy Gain increases the explainability of the model by providing insight to the model attention. Finally, the results of this method were analyzed to explain how the deep NLP model achieved a high classification performance. / Dissertation / Master of Applied Science (MASc) / In this document, an Artificial Intelligence (AI) approach for predicting 6-month Emergency Department (ED) visits is proposed. In this approach, the questionnaires gathered from children and youth admitted to an outpatient or inpatient clinic are converted to a text representation called Textionnaire. Next, AI is utilized to analyze the Textionnaire and predict the possibility of a future ED visit. This method was successful in about 75% of the time. In addition to the AI solution, an explainability component is introduced to explain how the natural language processing algorithm identifies the high risk patients.

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.

Here, we show that training with multiple noncommunicable diseases (NCDs) is both feasible and beneficial to modeling this class of diseases. We first use data from the Cancer Genome Atlas (TCGA) to train a pan cancer model, and then characterize the information the model has learned about the cancers. In doing this we show that the model has learned concepts that are relevant to the task of cancer classification. We also test the model on datasets derived independently of the TCGA cohort and show that the model is robust to data outside of its training distribution such as precancerous legions and metastatic samples. We then utilize the cancer model as the basis of a transfer learning study where we retrain it on other, non-cancer NCDs. In doing so we show that NCDs with very differing underlying biology contain extractible information relevant to each other allowing for a broader model of NCDs to be developed with existing datasets. We then test the importance of the samples source tissue in the model and find that the NCD class and tissue source may not be independent in our model. To address this, we use the tissue encodings to create augmented samples. We test how successfully we can use these augmented samples to remove or diminish tissue source importance to NCD class through retraining the model. In doing this we make key observations about the nature of concept importance and its usefulness in future neural network explainability efforts.

Nya neurala modeller har lett till stora framsteg inom maskinöversättning, men fungerar fortfarande sämre på språk som saknar stora mängder parallella data, så kallade lågresursspråk. Gawarbati är ett litet, hotat lågresursspråk där endast 5000 parallella meningar finns tillgängligt. Denna uppsats använder överföringsinlärning och hyperparametrar optimerade för små datamängder för att undersöka möjligheter och begränsningar för neural maskinöversättning från gawarbati till engelska. Genom att använda överföringsinlärning där en föräldramodell först tränades på hindi-engelska förbättrades översättningar med 1.8 BLEU och 1.3 chrF. Hyperparametrar optimerade för små datamängder ökade BLEU med 0.6 men minskade chrF med 1. Att kombinera överföringsinlärning och hyperparametrar optimerade för små datamängder försämrade resultatet med 0.5 BLEU och 2.2 chrF. De neurala modellerna jämförs med och presterar bättre än ordbaserad statistisk maskinöversättning och GPT-3. Den bäst presterande modellen uppnådde endast 2.8 BLEU och 19 chrF, vilket belyser begränsningarna av maskinöversättning på lågresursspråk samt det kritiska behovet av mer data. / Recent neural models have led to huge improvements in machine translation, but performance is still suboptimal for languages without large parallel datasets, so called low resource languages. Gawarbati is a small, threatened low resource language with only 5000 parallel sentences. This thesis uses transfer learning and hyperparameters optimized for small datasets to explore possibilities and limitations for neural machine translation from Gawarbati to English. Transfer learning, where the parent model was trained on parallel data between Hindi and English, improved results by 1.8 BLEU and 1.3 chrF. Hyperparameters optimized for small datasets increased BLEU by 0.6 but decreased chrF by 1. Combining transfer learning and hyperparameters optimized for small datasets led to a decrease in performance by 0.5 BLEU and 2.2 chrF. The neural models outperform a word based statistical machine translation and GPT-3. The highest performing model only achieved 2.8 BLEU and 19 chrF, which illustrates the limitations of machine translation for low resource languages and the critical need for more data. / VR 2020-01500

Le processus du don d’organes, crucial pour la survie de nombreux patients, ne répond pas à la demande croissante. Il dépend d’une identification, par les cliniciens, des potentiels donneurs d’organes. Cette étape est imparfaite et manque entre 30% et 60% des potentiels donneurs d’organes et ce indépendamment des pays étudiés. Améliorer ce processus est un impératif à la fois moral et économique. L’objectif de ce mémoire était de développer et valider un modèle afin de détecter automatiquement les potentiels donneurs d’organes. Pour ce faire, les données cliniques de l’ensemble des patients adultes hospitalisés aux soins intensifs du CHUM entre 2012 et 2019 ont été utilisées. 103 valeurs de laboratoires temporelles différentes et 2 valeurs statiques ont été utilisées pour développer un modèle de réseaux de neurones convolutifs entrainé à prédire les potentiels donneurs d’organes. Ce modèle a été comparé à un modèle fréquentiste linéaire non temporel. Le modèle a par la suite été validé dans une population externe cliniquement distincte. Différentes stratégies ont été comparées pour peaufiner le modèle dans cette population externe et améliorer les performances. Un total de 19 463 patients, dont 397 donneurs potentiels, ont été utilisés pour développer le modèle et 4 669, dont 36 donneurs potentiels, ont été utilisés pour la validation externe. Le modèle démontrait une aire sous la courbe ROC (AUROC) de 0.966 (IC95% 0.9490.981), supérieure au modèle fréquentiste linéaire (AUROC de 0.940 IC95% 0.908-0.969, p=0.014). Le modèle était aussi supérieur dans certaines sous populations d’intérêt clinique. Dans le groupe de validation externe, l’AUROC du modèle de réseaux de neurones était de 0.820 (0.682-0.948) augmentant à 0.874 (0.731-0.974) à l’aide d’un ré-entrainement. Ce modèle prometteur a le potentiel de modifier et d’améliorer la détection des potentiels donneurs d’organes. D’autres étapes de validation prospectives et d’amélioration du modèle, notamment l’ajout de données spécifiques, sont nécessaires avant une utilisation clinique de routine. / The organ donation process, however crucial for many patients’ survival, is not enough to address the increasing demand. Its efficiency depends on potential organ donors’ identification by clinicians. This imperfect step misses between 30%–60% of potential organ donor. Improving that process is a moral and economic imperative. The main goal of this work was to address that liming step by developing and validating a predictive model that could automatically detect potential organ donors. The clinical data from all patients hospitalized, between 2012 and 2019 to the CHUM critical care units were extracted. The temporal evolution of 103 types of laboratory analysis and 2 static clinical data was used to develop and test a convolutive neural network (CNN), trained to predict potential organ donors. This model was compared to a non-temporal logistical model as a baseline. The CNN model was validated in a clinically distinct external population. To improve the performance in this external cohort, strategies to fine-tune the network were compared. 19 463 patients, including 397 potential organ donors, were used to create the model and 4 669 patients, including 36 potential organ donors, served as the external validation cohort. The CNN model performed better with an AUROC of 0.966 (IC95% 0.949-0.981), compared to the logistical model (AUROC de 0.940 IC95% 0.908-0.969, p=0.014). The CNN model was also superior in specific subpopulation of increased clinical interest. In the external validation cohort, the CNN model’s AUROC was 0.820 (0.682-0.948) and could be improved to 0.874 (0.731-0.974) after fine tuning. This promising model could change potential organ donors' detection for the better. More studies are however required to improve the model, by adding more types of data, and to validate prospectively the mode before routine clinical usage.

The struggle of training adequate MT models using data-hungry NMT frameworks for low-resource language pairs has created a need to alleviate the scarcity of sufficiently large parallel corpora. Different transfer learning methods have been introduced as possible solutions to this problem, where a new model for a target task is initialized using parameters learned from some other high-resource task. Many of these methods are claimed to increase the translation quality of NMT systems in some low-resource environments, however, they are often proven to do so using different parent and child language pairs, a variation in data size, NMT frameworks, and training hyperparameters, which makes comparing them impossible. In this thesis project, three such transfer learning methods are put head-to-head in a controlled environment where the target task is to translate from the under-resourced Baltic languages Lithuanian and Latvian to English. In this controlled environment, the same parent language pairs, data sizes, data domains, transformer framework, and training parameters are used to ensure fair comparisons between the three transfer learning methods. The experiments involve training and testing models using all different combinations of transfer learning methods, parent language pairs, and either in-domain or out-domain data for an extensive study where different strengths and weaknesses are observed. The results display that Multi-Round Transfer Learning improves the overall translation quality the most but, at the same time, requires the longest training time by far. The Parameter freezing method provides a marginally lower overall improvement of translation quality but requires only half the training time, while Trivial Transfer learning improves quality the least. Both Polish and Russian work well as parents for the Baltic languages, while web-crawled data improves out-domain translations the most. The results suggest that all transfer learning methods are effective in a simulated low-resource environment, however, none of them can compete with simply having a larger target language pair data set, due to none of them overcoming the strong higher-resource baseline.

The aim of this paper is first of all to determine whether deep learning methods can recover trading strategies based on historical price and volume data, with scarcity of real data in mind. The second aim is to evaluate the methods to generate a deep learning blueprint for strategy extraction. Trading strategies can be built on many different types of data, often combined from different areas. In this paper, we focus on trading strategies based solely on historical price and volume data to limit the scope of the problem. Combinations of different deep learning architectures and methods such as transfer- and ensemble methods were evaluated. The results clearly show that deep learning models can recover relatively complex trading strategies to some extent. Models leveraging transfer learning outperform other models when data is scarce and ensemble methods elevate performance in certain regards. / Målet med denna rapport är i första hand att ta reda på om djupinlärningsmetoder kan återskapa handlingsstragetier baserat på historiska priser och volymdata, med vetskapen att datan är begränsad. Det andra målet är att utvärdera metoder för att skapa en djupinlärningsmall för att utvinna handelsstrategier. Handelsstrategier kan vara byggda på många olika datatyper, ofta i kombination från olika områden. I denna rapport fokuserar vi på strategier som enbart är baserade på historiska priser och volymdata för att begränsa problemet. Kombinationer av olika djupinlärningsarkitekturer tillsammans med metoder som till exempel överföringsinlärning och ensembleinlärning utvärderades. Resultaten visar tydligt att djupinlärningsmodeller kan återskapa relativt komplexa handlingsstrategier. Modeller som utnyttjade överföringsinlärning presterade bättre än andra modeller när datan var begränsad och ensembleinlärning ökade prestandan ytterligare i vissa sammanhang. / Kandidatexjobb i elektroteknik 2022, KTH, Stockholm

Subsurface porous media, like aquifers, petroleum reservoirs, and geothermal systems, are vital for natural resources and environmental management. Extensive research has been conducted to understand flow and transport in these media, addressing challenges in hydrocarbon extraction, carbon storage and waste management. Classifying the type of porous media (e.g., sandstone, carbonate) is often the first step in the rock characterization process, and it provides critical information regarding the physical properties of the porous media. Therefore, we utilize multivariate statistical methods with discriminant analysis to categorize porous media samples which proved to be efficient by achieving excellent classification accuracy on testing datasets and served as a surrogate tool to study key porous media characteristics. While recent advances in three-dimensional (3D) imaging of core samples have enabled digital subsurface characterization, the exorbitant computational cost associated with direct numerical simulation in 3D remains a persistent challenge. In contrast, machine learning (ML) models are much more efficient, though their use in subsurface characterization is still in its infancy. Therefore, we introduce a novel 3D convolution neural network (CNN) for end-to-end prediction of permeability. By increasing dataset size, diversity, and optimizing the network architecture, our model surpasses the accuracy of existing 3D CNN models for permeability prediction. It demonstrates excellent generalizability, accurately predicting permeability in previously unseen samples. However, despite the efficiency of the developed 3D CNN model for accurate and fast permeability prediction, its utility remains limited to small subdomains of the digital rock samples. Therefore, we introduce an upscaling technique using a new analytical solution to calculate effective permeability in a 3D digital rock composed of 2 × 2 × 2 anisotropic cells. By incorporating this solution into physics-informed neural network (PINN) models, we achieve highly accurate results. Even when upscaling previously unseen samples at multiple levels, the PINN with the physics-informed module maintains excellent accuracy. This advancement enhances the capability of ML models, like 3D CNN, for efficient and accurate digital rock analysis at the core scale. After successfully applying ML models in permeability prediction, we now extend their application to another important parameter in subsurface engineering projects: effective thermal conductivity, which is a key parameter in engineering projects like radioactive waste repositories, geothermal energy production, and underground energy storage. To address the need for large training data and processing power in ML models, we propose a novel framework based on transfer learning. This approach allows prior knowledge from previous applications to be transferred, resulting in faster and more efficient implementation of new relevant applications. We introduce CNN models trained on various porous media samples that leverage transfer learning to predict porous media sample thermal conductivity accurately. Our approach reduces training time, processing power, and data requirements, enabling effective prediction and analysis of porous media properties such as permeability and thermal conductivity. It also facilitates the application of ML to other properties, improving efficiency and accuracy. / Thesis / Doctor of Philosophy (PhD)

This abstract presents a study in which knowledge distillation techniques were applied to a Large Language Model (LLM) to create smaller, more efficient models without sacrificing performance. Three configurations of the RoBERTa model were selected as ”student” models to gain knowledge from a pre-trained ”teacher” model. Multiple steps were used to improve the knowledge distillation process, such as copying some weights from the teacher to the student model and defining a custom loss function. The selected task for the knowledge distillation process was sentiment analysis on Amazon Reviews for Sentiment Analysis dataset. The resulting student models showed promising performance on the sentiment analysis task capturing sentiment-related information from text. The smallest of the student models managed to obtain 98% of the performance of the teacher model while being 45% lighter and taking less than a third of the time to analyze an entire the entire IMDB Dataset of 50K Movie Reviews dataset. However, the student models struggled to produce meaningful results on the topic modeling task. These results were consistent with the topic modeling results from the teacher model. In conclusion, the study showcases the efficacy of knowledge distillation techniques in enhancing the performance of LLMs on specific downstream tasks. While the model excelled in sentiment analysis, further improvements are needed to achieve desirable outcomes in topic modeling. These findings highlight the complexity of language understanding tasks and emphasize the importance of ongoing research and development to further advance the capabilities of NLP models. / Denna sammanfattning presenterar en studie där kunskapsdestilleringstekniker tillämpades på en stor språkmodell (Large Language Model, LLM) för att skapa mindre och mer effektiva modeller utan att kompremissa på prestandan. Tre konfigurationer av RoBERTa-modellen valdes som ”student”-modeller för att inhämta kunskap från en förtränad ”teacher”-modell. Studien mäter även modellernas prestanda på två ”DOWNSTREAM” uppgifter, sentimentanalys och ämnesmodellering. Flera steg användes för att förbättra kunskapsdestilleringsprocessen, såsom att kopiera vissa vikter från lärarmodellen till studentmodellen och definiera en anpassad förlustfunktion. Uppgiften som valdes för kunskapsdestilleringen var sentimentanalys på datamängden Amazon Reviews for Sentiment Analysis. De resulterande studentmodellerna visade lovande prestanda på sentimentanalysuppgiften genom att fånga upp information relaterad till sentiment från texten. Den minsta av studentmodellerna lyckades erhålla 98% av prestandan hos lärarmodellen samtidigt som den var 45% lättare och tog mindre än en tredjedel av tiden att analysera hela IMDB Dataset of 50K Movie Reviews datasettet.Dock hade studentmodellerna svårt att producera meningsfulla resultat på ämnesmodelleringsuppgiften. Dessa resultat överensstämde med ämnesmodelleringsresultaten från lärarmodellen. Dock hade studentmodellerna svårt att producera meningsfulla resultat på ämnesmodelleringsuppgiften. Dessa resultat överensstämde med ämnesmodelleringsresultaten från lärarmodellen.

[pt] As Redes Generativas Adversariais (GANs) são modelos não supervisionados capazes de aprender a partir de um número indefinidamente grande de imagens. Entretanto, modelos que geram imagens a partir de linguagem dependem de dados rotulados de alta qualidade, que são escassos. A transferência de aprendizado é uma técnica conhecida que alivia a necessidade de dados rotulados, embora transformar um modelo gerativo incondicional em um modelo condicionado a texto não seja uma tarefa trivial. Este trabalho propõe uma abordagem de ajuste simples, porém eficaz, chamada U2C transfer. Esta abordagem é capaz de aproveitar modelos pré-treinados não condicionados enquanto aprende a respeitar as condições textuais fornecidas. Avaliamos a eficiência do U2C transfer ao ajustar o StyleGAN2 em duas das fontes de dados mais utilizadas para a geração images a partir de texto, resultando na arquitetura Text-Conditioned StyleGAN2 (TC-StyleGAN2). Nossos modelos alcançaram rapidamente o estado da arte nas bases de dados CUB-200 e Oxford-102, com valores de FID de 7.49 e 9.47, respectivamente. Esses valores representam ganhos relativos de 7 por cento e 68 por cento, respectivamente, em comparação com trabalhos anteriores. Demonstramos que nosso método é capaz de aprender detalhes refinados a partir de consultas de texto, produzindo imagens fotorrealistas e detalhadas. Além disso, mostramos que os modelos organizam o espaço intermediário de maneira semanticamente significativa. Nossas descobertas revelam que as imagens sintetizadas usando nossa técnica proposta não são apenas críveis, mas também exibem forte alinhamento com suas descrições textuais correspondentes. De fato, os escores de alinhamento textual alcançados por nosso método são impressionantemente e comparáveis aos das imagens reais. / [en] Generative Adversarial Networks (GANs) are unsupervised models that can learn from an indefinitely large amount of images. On the other hand, models that generate images from language queries depend on high-quality labeled data that is scarce. Transfer learning is a known technique that alleviates the need for labeled data, though it is not trivial to turn an unconditional generative model into a text-conditioned one. This work proposes a simple, yet effective fine-tuning approach, called Unconditional-to-Conditional Transfer Learning (U2C transfer). It can leverage well-established pre-trained models while learning to respect the given textual condition conditions. We evaluate U2C transfer efficiency by fine-tuning StyleGAN2 in two of the most widely used text-to-image data sources, generating the Text-Conditioned StyleGAN2 (TC-StyleGAN2). Our models quickly achieved state-of-the-art results in the CUB-200 and Oxford-102 datasets, with FID values of 7.49 and 9.47, respectively. These values represent relative gains of 7 percent and 68 percent compared to prior work. We show that our method is capable of learning fine-grained details from text queries while producing photorealistic and detailed images. Our findings highlight that the images created using our proposed technique are credible and display a robust alignment with their corresponding textual descriptions.