Semantische Suche

To accomplish this, we will analyze and summarize the prevalent research related to microblog data-based forecasting and analysis, with a focus on the data processing and mining approach. Based on the findings, one or several candidate frameworks are developed and evaluated by testing the correlation of their generated data sets against the target time series they are generated for.

While summative research on microblog data-based correlation analysis exists, it is mainly focused on summarizing the state of the field. This thesis adds to the body of research by applying summarized findings and generating experimental evidence regarding the generalizability of microblog data mining approaches and their effectiveness.

To define “change points”, we assume that there is a distribution, which may change over time, generating the data we observe. A change point then is a change in this underlying distribution, i.e., the distribution coming before a change point is different from the distribution coming after. The principled way to compare distributions, and to find change points, is to employ statistical tests.

While change point detection is an unsupervised problem in practice, i.e., the data is unlabelled, the development and evaluation of data analysis algorithms requires labelled data. Only few labelled real world data sets are publicly available and many of them are either too small or have ambiguous labels. Further issues are that reusing data sets may lead to overfitting, and preprocessing (e.g., removing outliers) may manipulate results. To address these issues, van den Burg et al. publish 37 data sets annotated by data scientists and ML researchers and use them for an assessment of 14 change detection algorithms. Yet, there remain concerns due to the fact that these are labelled by hand: Can humans correctly identify changes according to the definition, and can they be consistent in doing so?

The goal of this Bachelor's thesis is to algorithmically label their data sets following the formal definition and to also identify and label larger and higher-dimensional data sets, thereby extending their work. To this end, we leverage a non-parametric hypothesis test which builds on Maximum Mean Discrepancy (MMD) as a test statistic, i.e., we identify changes in a principled way. We will analyse the labels so obtained and compare them to the human annotations, measuring their consistency with the F1 score. To assess the influence of the algorithmic and definition-conform annotations, we will use them to reevaluate the algorithms of van den Burg et al. and compare the respective performances.

We extend the recently introduced methodology of Monte Carlo Dependency Estimation (MCDE), an effective and efficient traditional multivariate dependency estimator. We introduce Generalized Monte Carlo Dependency Estimation (gMCDE) and focus in particular on the highly relevant subproblem of generalized dependency estimation, known as canonical dependency estimation, which aims to estimate the dependency between two multidimensional datasets. We demonstrate the practical relevance of Canonical Monte Carlo Dependency Estimation (cMCDE) by applying it to feature selection, introducing two methodologies for anytime supervised filter feature selection, Canonical Monte Carlo Feature Selection (cMCFS) and Canonical Multi Armed Bandit Feature Selection (cMABFS). cMCFS directly applies the methodology of cMCDE to feature selection, while cMABFS treats the feature selection problem as a multi armed bandit problem, which utilizes cMCDE to determine relevant features.

In this thesis, we aim to develop a faster injection molding simulation based on Graph Neural Networks (GNNs) as a surrogate model. Our approach learns a simulation as a composition of three functions: an encoder, a processor and a decoder. The encoder takes in a graph representation of a 3D geometry of an injection molding part and returns a numeric embedding of each node in the graph. The processor updates the embeddings of each node multiple times based on its neighbors. The decoder then decodes the final embeddings of each node into physically meaningful variables, say, the fill state of the node. Our model can predict the progression of the flow front during a time step with a fixed size. To simulate a full mold filling process, our model is applied sequentially until the entire mold is filled. Our architecture is applicable to any kind of material, geometry and injection process parameters. We evaluate our architecture by its accuracy and runtime when predicting node properties. We also evaluate our models transfer learning ability on a real world injection molding part.