No description available.

[pt] Tabelas são uma forma bastante comum de organizar e publicar dados. Por exemplo, a Web possui um enorme número de tabelas publicadas em HTML, embutidas em documentos em PDF, ou que podem ser simplesmente baixadas de páginas Web. Porém, tabelas nem sempre são fáceis de interpretar pois possuem uma grande variedade de características e são organizadas de diversas formas. De fato, um grande número de métodos e ferramentas foram desenvolvidos para interpretação de tabelas. Esta dissertação apresenta a implementação de um algoritmo, baseado em Conditional Random Fields (CRFs), para classificar as linhas de uma tabela em linhas de cabeçalho, linhas de dados e linhas de metadados. A implementação é complementada por dois algoritmos para reconhecimento de tabelas em planilhas, respectivamente baseados em regras e detecção de regiões. Por fim, a dissertação descreve os resultados e os benefícios obtidos pela aplicação dos algoritmos a tabelas em formato HTML, obtidas da Web, e a tabelas em forma de planilhas, baixadas do Web site da Agência Nacional de Petróleo. / [en] Tables are widely adopted to organize and publish data. For example, the Web has an enormous number of tables, published in HTML, imbedded in PDF documents, or that can be simply downloaded from Web pages. However, tables are not always easy to interpret because of the variety of features and formats used. Indeed, a large number of methods and tools have been developed to interpret tables. This dissertation presents the implementation of an algorithm, based on Conditional Random Fields (CRFs), to classify the rows of a table as header rows, data rows or metadata rows. The implementation is complemented by two algorithms for table recognition in a spreadsheet document, respectively based on rules and on region detection. Finally, the dissertation describes the results and the benefits obtained by applying the implemented algorithms to HTML tables, obtained from the Web, and to spreadsheet tables, downloaded from the Brazilian National Petroleum Agency.

Machine Learning has become more important for materials engineering in the last decade. Globally, automated machine learning (AutoML) is growing in popularity with the increasing demand for data analysis solutions. Yet, it is not frequently used for small tabular data. Comparisons and benchmarks already exist to assess the qualities of AutoML tools in general, but none of them elaborates on the surrounding conditions of materials engineers working with experimental data: small datasets with less than 1000 samples. This benchmark addresses these conditions and draws special attention to the overall competitiveness with manual data analysis. Four representative AutoML frameworks are used to evaluate twelve domain-specific datasets to provide orientation on the promises of AutoML in the field of materials engineering. Performance, robustness and usability are discussed in particular. The results lead to two main conclusions: First, AutoML is highly competitive with manual model optimization, even with little training time. Second, the data sampling for train and test data is of crucial importance for reliable results.

This thesis investigates the effectiveness of Tabular Variational Auto Encoders (TVAEs) in generating high-quality synthetic tabular data and assesses their compliance with differential privacy principles. The study shows that while TVAEs are better than VAEs at generating synthetic data that faithfully reproduces the distribution of real data as measured by the Synthetic Data Vault (SDV) metrics, the latter does not guarantee that the synthetic data is up to the task in practical industrial applications. In particular, models trained on TVAE-generated data from the Creditcards dataset are ineffective. The author also explores various optimisation methods on TVAE, such as Gumbel Max Trick, Drop Out (DO) and Batch Normalization, while pointing out that techniques frequently used to improve two-dimensional TVAE, such as Kullback–Leibler Warm-Up and B Disentanglement, are not directly transferable to the one-dimensional context. However, differential privacy to TVAE was not implemented due to time constraints and inconclusive results. The study nevertheless highlights the benefits of stabilising training with the Differential Privacy - Stochastic Gradient Descent (DP-SGD), as with a dropout, and the existence of an optimal equilibrium point between the constraints of differential privacy and the number of training epochs in the model. / Denna avhandling undersöker hur effektiva Tabular Variational AutoEncoders (TVAE) är när det gäller att generera högkvalitativa syntetiska tabelldata och utvärderar deras överensstämmelse med differentierade integritetsprinciper. Studien visar att även om TVAE är bättre än VAE på att generera syntetiska data som troget återger fördelningen av verkliga data mätt med Synthetic Data Vault (SDV), garanterar det senare inte att de syntetiska data är upp till uppgiften i praktiska industriella tillämpningar. I synnerhet är modeller som tränats på TVAE-genererade data från Creditcards-datasetet ineffektiva. Författaren undersöker också olika optimeringsmetoder för TVAE, såsom Gumbel Max Trick, DO och Batch Normalization, samtidigt som han påpekar att tekniker som ofta används för att förbättra tvådimensionell TVAE, såsom Kullback-Leibler Warm-Up och B Disentanglement, inte är direkt överförbara till det endimensionella sammanhanget. På grund av tidsbegränsningar och redan ofullständiga resultat implementerades dock inte differentierad integritet för TVAE. Studien belyser ändå fördelarna med att stabilisera träningen med Differential Privacy - Stochastic Gradient Descent (DP-SGD), som med en drop-out, och förekomsten av en optimal jämviktspunkt mellan begränsningarna för differential privacy och antalet träningsepoker i modellen.

Artificial neural networks are some of the most powerful machine learning models and have gained interest in the telecommunications domain as well as other fields and applications due to their strong performance and flexibility. Creating these models typically requires manually choosing their architecture along with other hyperparameters that are crucial for their performance. Neural Architecture Search (NAS) seeks to automate architecture choice and has gained increasing interest in recent years. In this thesis, we propose a new NAS method based on differential architecture search (DARTS) to find architectures of fully-connected feed forward networks on tabular datasets. We train a gating mechanism on a validation dataset and compare four candidate gate functions as a tool to determine the number of hidden units per hidden layer in our neural networks for different tasks. Our findings show that our new method can reliably find architectures that are more compact and outperform manually chosen architectures. Interestingly, we also found that extracting weights learned during the search process could generate models that achieve significantly higher and more stable performance than identical architectures retrained from scratch. Our method achieved equal in performance to that of another NAS-method, while only requiring half an hour of training compared to 280 hours. The trained models also demonstrated a competitive performance when benchmarked to other state-of-the-art machine learning models. The primary benefit of our method, stems from the extraction and fine-tuning of certain weights. Our results indicate that improvements from extracted weights could relate to the lottery ticket hypothesis of neural networks, which invites further study for a fuller understanding.

Digitalization has led to tons of available customer data and possibilities for data-driven innovation. However, the data needs to be handled carefully to protect the privacy of the customers. Generative Adversarial Networks (GANs) are a promising recent development in generative modeling. They can be used to create synthetic data which facilitate analysis while ensuring that customer privacy is maintained. Prior research on GANs has shown impressive results on image data. In this thesis, we investigate the viability of using GANs within the financial industry. We investigate two state-of-the-art GAN models for synthesizing tabular data, TGAN and CTGAN, along with a simpler GAN model that we call WGAN. A comprehensive evaluation framework is developed to facilitate comparison of the synthetic datasets. The results indicate that GANs are able to generate quality synthetic datasets that preserve the statistical properties of the underlying data and enable a viable and reproducible subsequent analysis. It was however found that all of the investigated models had problems with reproducing numerical data. / Digitaliseringen har fört med sig stora mängder tillgänglig kunddata och skapat möjligheter för datadriven innovation. För att skydda kundernas integritet måste dock uppgifterna hanteras varsamt. Generativa Motstidande Nätverk (GANs) är en ny lovande utveckling inom generativ modellering. De kan användas till att syntetisera data som underlättar dataanalys samt bevarar kundernas integritet. Tidigare forskning på GANs har visat lovande resultat på bilddata. I det här examensarbetet undersöker vi gångbarheten av GANs inom finansbranchen. Vi undersöker två framstående GANs designade för att syntetisera tabelldata, TGAN och CTGAN, samt en enklare GAN modell som vi kallar för WGAN. Ett omfattande ramverk för att utvärdera syntetiska dataset utvecklas för att möjliggöra jämförelse mellan olika GANs. Resultaten indikerar att GANs klarar av att syntetisera högkvalitativa dataset som bevarar de statistiska egenskaperna hos det underliggande datat, vilket möjliggör en gångbar och reproducerbar efterföljande analys. Alla modellerna som testades uppvisade dock problem med att återskapa numerisk data.

The development of new solutions incorporating artificial intelligence (AI) within the medical field is an area of great interest. However, access to comprehensive and diverse datasets is restricted due to the sensitive nature of the data. A potential solution to this is to generatesynthetic datasets based on real medical data. Synthetic data could protect the integrity of the subjects while preserving the inherent information necessary for training AI models and be generated in greater quantity than otherwise available. This thesis project aims to generate reliable clinical trial subject data using a generative adversarial network (GAN). The main data set used is a mock clinical trial dataset consisting of multiple subject visits, however an additional data set containing authentic medical data is also used for better insights into the model’s ability to learn underlying relationships. The thesis also investigates training strategies for simulating the temporal dimension and the missing values in the data. The GAN model used is an altered version of the Conditional Tabular GAN (CTGAN)made to be compatible with the preprocessed clinical trial mock data, and multiple model architectures and number of training epochs are examined. The results show great potential for GAN models on clinical trial datasets, especially for real-life data. One model, trained on the authentic dataset, generates near-perfect synthetic data with respect to column distributions and correlation between columns. The results also show that classification models trained on synthetic data and tested on real data have the potential to match the performance of classification models trained on real data. While the synthetic data replicates the missing values, no definitive conclusion can be drawn regarding the temporal characteristics due to the sparsity of the mock dataset and lack of real correlations in it. Although the results are promising, further experiments on authentic datasets with less sparsity are required.

<p>Disertacija se bavi analizom uticaja raspodela podataka na rezultate algoritama diskretizacije u okviru procesa ma&scaron;inskog učenja. Na osnovu izabranih baza i algoritama diskretizacije teorije grubih skupova i stabala odlučivanja, istražen je uticaj odnosa raspodela podataka i tačaka reza određene diskretizacije.<br />Praćena je promena konzistentnosti diskretizovane tabele u zavisnosti od položaja redukovane tačke reza na histogramu. Definisane su fiksne tačke reza u zavisnosti od segmentacije multimodal raspodele, na osnovu kojih je moguće raditi redukciju preostalih tačaka reza. Za određivanje fiksnih tačaka konstruisan je algoritam FixedPoints koji ih određuje u skladu sa grubom segmentacijom multimodal raspodele.<br />Konstruisan je algoritam aproksimativne diskretizacije APPROX MD za redukciju tačaka reza, koji koristi tačke reza dobijene algoritmom maksimalne razberivosti i parametre vezane za procenat nepreciznih pravila, ukupni procenat klasifikacije i broj tačaka redukcije. Algoritam je kompariran u odnosu na algoritam maksimalne razberivosti i u odnosu na algoritam maksimalne razberivosti sa aproksimativnim re&scaron;enjima za &alpha;=0,95.</p> / <p>This dissertation analyses the influence of data distribution on the results of discretization algorithms within the process of machine learning. Based on the chosen databases and the discretization algorithms within the rough set theory and decision trees, the influence of the data distribution-cuts relation within certain discretization has been researched.<br />Changes in consistency of a discretized table, as dependent on the position of the reduced cut on the histogram, has been monitored. Fixed cuts have been defined, as dependent on the multimodal segmentation, on basis of which it is possible to do the reduction of the remaining cuts. To determine the fixed cuts, an algorithm FixedPoints has been constructed, determining these points in accordance with the rough segmentation of multimodal distribution.<br />An algorithm for approximate discretization, APPROX MD, has been constructed for cuts reduction, using cuts obtained through the maximum discernibility (MD-Heuristic) algorithm and the parametres related to the percent of imprecise rules, the total classification percent and the number of reduction cuts. The algorithm has been compared to the MD algorithm and to the MD algorithm with approximate solutions for &alpha;=0,95.</p>

The flight search engine is a technology used in the air travel industry. It allows the traveler to search and book for the best flight options, such as the combination of flights while keeping the best services, options, and price. The computation for a flight search query can be very intensive given its parameters and complexity. The project goal is to predict the flight search queries computation cost for a new flight search engine product when dealing with parameters change and optimizations. The problem of flight search cost prediction is a regression problem. We propose to solve the problem by delimiting the problem based on its business logic and meaning. Our problem has data defined as a graph, which is why we have chosen Graph Neural Network. We have investigated multiple pretraining strategies for the evaluation of node embedding concerning a realworld regression task, including using a line graph for the training. The embeddings are used for downstream regression tasks. Our work is based on some stateoftheart Machine Learning, Deep Learning, and Graph Neural Network methods. We conclude that for some business use cases, the predictions are suitable for production use. In addition, the prediction of tree ensemble boosting methods produces negatives predictions which further degrade the R2 score by 4% because of the business meaning. The Deep Neural Network outperformed the most performing Machine Learning methods by 8% to 12% of R2 score. The Deep Neural Network also outperformed Deep Neural Network with pretrained node embedding from the Graph Neural Network methods by 11% to 17% R2 score. The Deep Neural Network achieved 93%, 81%, and 63% R2 score for each task with increasing difficulty. The training time range from 1 hour for Machine Learning models, 2 to 10 hours for Deep Learning models, and 8 to 24 hours for Deep Learning model for tabular data trained end to end with Graph Neural Network layers. The inference time is around 15 minutes. Finally, we found that using Graph Neural Network for the node regression task does not outperform Deep Neural Network. / Flygsökmotor är en teknik som används inom flygresebranschen. Den gör det möjligt för resenären att söka och boka de bästa flygalternativen, t.ex. kombinationer av flygningar med bästa service, alternativ och pris. Beräkningen av en flygsökning kan vara mycket intensiv med tanke på dess parametrar och komplexitet. Projektets mål är att förutsäga beräkningskostnaden för flygsökfrågor för en ny produkt för flygsökmotor när parametrar ändras och optimeringar görs. Problemet med att förutsäga kostnaderna för flygsökning är ett regressionsproblem. Vi föreslår att man löser problemet genom att avgränsa det utifrån dess affärslogik och innebörd. Vårt problem har data som definieras som en graf, vilket är anledningen till att vi har valt Graph Neural Network. Vi har undersökt flera förträningsstrategier för utvärdering av nodinbäddning när det gäller en regressionsuppgift från den verkliga världen, bland annat genom att använda ett linjediagram för träningen. Inbäddningarna används för regressionsuppgifter i efterföljande led. Vårt arbete bygger på några toppmoderna metoder för maskininlärning, djupinlärning och grafiska neurala nätverk. Vi drar slutsatsen att förutsägelserna är lämpliga för produktionsanvändning i vissa Vi drar slutsatsen att förutsägelserna är lämpliga för produktionsanvändning i vissa fall. Dessutom ger förutsägelserna från trädens ensemble av boostingmetoder negativa förutsägelser som ytterligare försämrar R2poängen med 4% på grund av affärsmässiga betydelser. Deep Neural Network överträffade de mest effektiva metoderna för maskininlärning med 812% av R2poängen. Det djupa neurala nätverket överträffade också det djupa neurala nätverket med förtränad node embedding från metoderna för grafiska neurala nätverk med 11 till 17% av R2poängen. Deep Neural Network uppnådde 93, 81 och 63% R2poäng för varje uppgift med stigande svårighetsgrad. Träningstiden varierar från 1 timme för maskininlärningsmodeller, 2 till 10 timmar för djupinlärningsmodeller och 8 till 24 timmar för djupinlärningsmodeller för tabelldata som tränats från början till slut med grafiska neurala nätverkslager. Inferenstiden är cirka 15 minuter. Slutligen fann vi  att användningen av Graph Neural Network för uppgiften om regression av noder inte överträffar Deep Neural Network.