Examensarbeten för masterexamen
Länka till denna samling:
Browse
Senast publicerade
Visar 1 - 5 av 416
- PostMultidimensional Data-Driven Modelling of Engine Test Cell Data(2021) Andersson, Helena; Chalmers tekniska högskola / Institutionen för matematiska vetenskaper; Sagitov, Serik; Andersson-Hedberg, Per; Johansson, AntonIn the journey towards a more sustainable vehicle fleet, requirements for lower emissions and improved energy efficiency in gasoline engines lead to more components being added to the internal combustion engines. This adds to the degrees of freedom when trying to model air flow in the engine using volumetric efficiency. This paper presents a way of modelling volumetric efficiency from engine test cell data provided by T-Engineering – a company that designs and develops control systems for vehicles. The model uses Gaussian process regression (GPR) for inter- and extrapolation, including noise reduction of the measurement data. Furthermore, a local interpretable model-agnostic explainer (LIME) is used to find regions of uncertainty by explaining what features contribute to increasing the variance of the GPR predictions. In addition, a neural network model is implemented in order to improve the prediction runtime, with the purpose of enabling real-time predictions in the control systems. The model(s) were found to give a more physically accurate description of volumetric efficiency than the one currently used at T-Engineering. The runtime for making predictions for 50 data points with the neural network was ~ 0.14 ms on an AMD Ryzen 7 PRO 4750U with Radeon Graphics 1.70 GHz and 32.0GB RAM. It remains to investigate what the runtime on a limited CPU in the control systems will be.
- PostThe abundance and selection of antibiotic resistance genes: A metagenomic study based on wastewater and human gut data.(2024) Holmström, Michaela; Chalmers tekniska högskola / Institutionen för matematiska vetenskaper; Kristiansson, Erik; Johnning, Anna; Lund, DavidAntibiotic resistance is a global issue, with many consequences for the individual person and society as a whole. Studying potential new genes conferring resistance (latent ARGs) and the flow of resistance genes through environments could be crucial for the prevention and control of ARGs. This project aimed firstly to investigate how antibiotic usage affects new antibiotic resistance genes in the human gut. A metagenomic pipeline including mapping of reads and statistical models using a study of 12 healthy individuals treated with three broad-spectrum antibiotics enabled analysis of ARGs and taxonomy abundances. It was shown that after treatment with the broad-spectrum antibiotics, several latent ARGs were more abundant in the short term, implying the potential selection of these ARGs. The change in ARGs can not be ruled out as being due to the recolonization of fast-growing bacterial species, normally carrying such genes, as also the taxonomy composition was largely affected. This effect was not seen on the last sampling day, after 180 days, in which only one single ARG was found at higher abundance. The short-term effect could, nevertheless, impact the spread of ARGs due to the potential for these ARGs to spread in the disrupted microbiota. Secondly, this project aimed to investigate the flow of ARGs between the human gut and wastewater. For this, a large number of metagenomic samples were used. Similarly to the first part, a metagenomic pipeline was utilized for ARG and taxonomy analysis. Contrary to previous findings, the results implied that ARGs flow from the human gut to the wastewater, while not the other way around. Furthermore, pathogenic presence was highest for ARGs found at high prevalence in both the human gut and wastewater. Interestingly, ARGs present only in wastewater were more prevalent in pathogens compared to ARGs present only in the human gut. This implies that the spread of ARGs to pathogens could be linked to presence in wastewater environments. Moreover, the taxonomy of the human gut and wastewater differed. The results of this project can inform several future research directions, from broadening the data of longitudinal studies to a deeper dive into the flow of ARGs.
- PostPredicting antibiotic resistance using fusion transformers(2024) Olsson, Jesper; Chalmers tekniska högskola / Institutionen för matematiska vetenskaper; Kristiansson, Erik; Johnning, Anna; Kristiansson, ErikAntimicrobial resistance threatens recent gains in global public health by making it more difficult to treat infections. Clinicians must administer treatments based on limited diagnostic information and increasing resistance complicates these decisions. This thesis project explores ways to support this process by developing a framework for training a transformer model using data fusion of patient and genotype data with phenotype data to make individualized predictions of antibiotic resistance in Escherichia coli on these multimodal data. To achieve this, the model was trained in two stages: first, the model was pre-trained on large volumes of unimodal data using masked language modeling to learn patterns within the modalities; and second, the model was fine-tuned on a small multimodal dataset to learn patterns across modalities. To evaluate pre-training strategies, the model was fine-tuned on two clinically relevant tasks and smaller training sets. To determine the value of introducing multimodality and the effect of genotype data availability on performance, the model was fine-tuned on varying levels of available genotype information. The results show that the model performs well on the fine-tuning tasks, that pretraining on unimodal data improves performance, and that the model can extrapolate well from small training sets and incomplete data. Therefore, it can be concluded that this work has achieved the aim of developing a model that can make accurate predictions based on limited diagnostic information. Importantly, large performance improvements were observed with increasing genotype data availability, especially on difficult antibiotics. Furthermore, the model was better able to utilize available genotype information when pre-trained. However, while no clear conclusion on the best pre-training strategy can be drawn from the results of this work, they indicate that using systematic class masking in pre-training yields the highest performance. Future research should further investigate the best strategy for pre-training the model, how the model utilizes genotype data to improve performance, and how genotype data affects performance on limited training data.
- PostModelling COVID-19 Individual Risks in Sweden Using Spatial Information, Statistics and Machine Learning(2024) Fu, Lukas; Chalmers tekniska högskola / Institutionen för matematiska vetenskaper; Gerlee, Philip; Cronie, Ottmar; Han, BinThe Covid-19 pandemic was a modern time pandemic that lasted a little over two years, and caused a severe social and economical disruption on a worldwide scale. Using data consisting of individual and DeSO covariates of the population of Sweden, sourced from Statistics Sweden and the Public Health Agency of Sweden, this project aims to model individual risks of Covid-19 using machine learning algorithms, and to extract information on feature importance from the fitted models. The models tested include logistic regression, random forest, support vector machines and neural network, and Shapley values were additionally evaluated for random forest in an attempt to gain more insight into the feature relation to the prediction. The logistic regression and random forest models both resulted in feature importances consisting of a mixture of individual and DeSO features, where features such as age, level of education, and living conditions for both the DeSO and the individual, along with income and occupation of the individual, showed high importance. Support vector machines and neural network models did not produce any useful results due to computational limitations. The large size of the data set was a consistent hindrance in this project, as many issues were caused by computational costs, and many of the improvements on optimization in this project are centered around handling these costs. Further research may entail in optimizing performances of presented or alternate models, but may also expand to more thoroughly analyse the spatial and temporal dependencies of disease cases. While the results of this project might not be particularly significant on its own, this project may still provide a basis for future developments in pandemic data analysis.
- PostEmergence of Agency from a Causal Perspective(2024) Ånestrand, Alvin; Chalmers tekniska högskola / Institutionen för matematiska vetenskaper; Lundh, Torbjörn; Häggström, Olle; Fox, James; Everitt, TomCausal models of agents and agentic behavior allows for safety analysis of machine learning systems. Understanding how goal-directed behavior emerges from adapting to an environment is however non-trivial. This thesis addresses the gap between theoretical models and real-world implementations of machine learning systems, though a framework that formalizes the connection between system dynamics and goal-driven behavior. This thesis introduces novel probabilistic graphical models for describing system dynamics involving learning agents, based on dynamic bayesian networks, which allows for a flexible representation of causal relationships in the training environment. To analyze goal-directed behavior that emerges from interacions between agents and the environment, the thesis also introduces temporally abstracted models. Such a model captures the dynamics of a system after the learning process has converged, derived from a model of the learning process. A temporally abstracted model describes potential outcomes involving equilibria between agents and the environment, and can under certain conditions be viewed as a model of goal-directed behavior in the system.