[pt] Métodos ensemble como random forest, boosting e bagging foram extensivamente estudados e provaram ter uma acurácia melhor do que usar apenas um preditor. Entretanto, a desvantagem é que os modelos obtidos utilizando esses métodos podem ser muito mais difíceis de serem interpretados do que por exemplo, uma árvore de decisão. Neste trabalho, nós abordamos o problema de construir uma árvore de decisão que aproximadamente reproduza um conjunto de árvores, explorando o tradeoff entre acurácia e interpretabilidade, que pode ser alcançado quando a reprodução exata do conjunto de árvores é relaxada. Primeiramente, nós formalizamos o problem de obter uma árvore de decisão de uma determinada profundidade que seja a mais aderente ao conjunto de árvores e propomos um algoritmo de programação dinâmica para resolver esse problema. Nós também provamos que a árvore de decisão obtida por esse procedimento satisfaz garantias de generalização relacionadas a generalização do modelo original de conjuntos de árvores, um elemento crucial para a efetividade dessa árvore de decisão em prática. Visto que a complexidade computacional do algoritmo de programação dinâmica é exponencial no número de features, nós propomos duas heurísticas para gerar árvores de uma determinada profundidade com boa aderência em relação ao conjunto de árvores. Por fim, nós conduzimos experimentos computacionais para avaliar os algoritmos propostos. Quando utilizados classificadores mais interpretáveis, os resultados indicam que em diversas situações a perda em acurácia é pequena ou inexistente: restrigindo a árvores de decisão de profundidade 6, nossos algoritmos produzem árvores que em média possuem acurácias que estão a 1 por cento (considerando o algoritmo de programção dinâmica) ou 2 por cento (considerando os algoritmos heurísticos) do conjunto original de árvores. / [en] Ensemble methods in machine learning such as random forest, boosting, and bagging have been thoroughly studied and proven to have better accuracy than using a single predictor. However, their drawback is that they give models that can be much harder to interpret than those given by, for example, decision trees. In this work, we approach in a principled way the problem of constructing a decision tree that approximately reproduces a tree ensemble, exploring the tradeoff between accuracy and interpretability that can be obtained once exact reproduction is relaxed. First, we formally define the problem of obtaining the decision tree of a given depth that is most adherent to a tree ensemble and give a Dynamic Programming algorithm for solving this problem. We also prove that the decision trees obtained by this procedure satisfy generalization guarantees related to the generalization of the original tree ensembles, a crucial element for their effectiveness in practice. Since the computational complexity of the Dynamic Programming algorithm is exponential in the number of features, we also design heuristics to compute trees of a given depth with good adherence to a tree ensemble. Finally, we conduct a comprehensive computational evaluation of the algorithms proposed. The results indicate that in many situations, there is little or no loss in accuracy in working more interpretable classifiers: even restricting to only depth-6 decision trees, our algorithms produce trees with average accuracies that are within 1 percent (for the Dynamic Programming algorithm) or 2 percent (heuristics) of the original random forest.

This thesis investigates how pre-trained contextualized language models can be adapted for multi-label text classification of Swedish news articles. Various classifiers are built on pre-trained BERT and ELECTRA models, exploring global and local classifier approaches. Furthermore, the effects of domain specialization, using additional metadata features and model compression are investigated. Several hundred thousand news articles are gathered to create unlabeled and labeled datasets for pre-training and fine-tuning, respectively. The findings show that a local classifier approach is superior to a global classifier approach and that BERT outperforms ELECTRA significantly. Notably, a baseline classifier built on SVMs yields competitive performance. The effect of further in-domain pre-training varies; ELECTRA’s performance improves while BERT’s is largely unaffected. It is found that utilizing metadata features in combination with text representations improves performance. Both BERT and ELECTRA exhibit robustness to quantization and pruning, allowing model sizes to be cut in half without any performance loss.

With the advancement of neural networks and deep learning, the complexity and size of models have increased exponentially. On the other hand, advancements of internet of things (IoT) and sensor technology have opened for many embedded machine learning applications and projects. In many of these applications, the hardware has some constraints in terms of computational and memory resources. The always increasing popularity of these applications, require shrinking and compressing neural networks in order to satisfy the requirements. The frameworks and algorithms governing the compression of a neural network are commonly referred to as pruning algorithms. In this project several pruning frameworks are applied to different neural network architectures to better understand their effect on the performance as well as the size of the model. Through experimental evaluations and analysis, this thesis provides insights into the benefits and trade-offs of pruning algorithms in terms of size and performance, shedding light on their practicality and suitability for embedded machine learning. The findings contribute to the development of more efficient and optimized neural networks for resource constrained hardware, in real-world IoT applications such as wearable technology. / Med framstegen inom neurala nätverk och djupinlärning har modellernas komplexitet och storlek ökat exponentiellt. Samtidigt har framsteg inom Internet of Things (IoT) och sensorteknik öppnat upp för många inbyggda maskininlärningsapplikationer och projekt. I många av dessa applikationer finns det begränsningar i hårdvaran avseende beräknings- och minnesresurser. Den ständigt ökande populariteten hos dessa applikationer kräver att neurala nätverk minskas och komprimeras för att uppfylla kraven. Ramverken och algoritmerna som styr komprimeringen av ett neuralt nätverk kallas vanligtvis för beskärningsalgoritmer. I detta projekt tillämpas flera beskärningsramverk på olika neurala nätverksarkitekturer för att bättre förstå deras effekt på prestanda och modellens storlek. Genom experimentella utvärderingar och analys ger denna avhandling insikter om fördelarna och avvägningarna med beskärningsalgoritmer vad gäller storlek och prestanda, och belyser deras praktiska användbarhet och lämplighet för inbyggd maskininlärning. Resultaten bidrar till utvecklingen av mer effektiva och optimerade neurala nätverk för resursbegränsad hårdvara i verkliga IoT-applikationer, såsom bärbar teknik.