Large scale semantic segmentation is considered as one of the fundamental tasks in 3D scene understanding. Point clouds provide a basic and rich geometric representation of scenes and tangible objects. Convolutional Neural Networks (CNNs) have demonstrated an impressive success in processing regular discrete data such as 2D images and 1D audio. However, CNNs do not directly generalize to point cloud processing due to their irregular and un-ordered nature. One way to extend CNNs to point cloud understanding is to derive an intermediate euclidean representation of a point cloud by projecting onto image domain, voxelizing, or treating points as vertices of an un-directed graph. Graph-CNNs (GCNs) have demonstrated to be a very promising solution for deep learning on irregular data such as social networks, biological systems, and recently point clouds. Early works in literature for graph based point networks relied on constructing dynamic graphs in the node feature space to define a convolution kernel. Later works constructed hierarchical static graphs in 3D space for an encoder-decoder framework inspired from image segmentation. This thesis takes a closer look at both dynamic and static graph neighborhoods of graph- based point networks for the task of semantic segmentation in order to: 1) discuss a potential cause for why going deep in dynamic GCNs does not necessarily lead to an improved performance, and 2) propose a new approach in treating points in a static graph neighborhood for an improved information aggregation. The proposed method leads to an efficient graph based 3D semantic segmentation network that is on par with current state-of-the-art methods on both indoor and outdoor scene semantic segmentation benchmarks such as S3DIS and Semantic3D.

Neste trabalho, aprofundamos o estudo sobre risco sistêmico via funções de agregação. Consideramos três carteiras diferentes como proxy para um sistema econômico, estas carteiras são consistidas por duas funções de agregação, baseadas em todos as ações do E.U.A, e um índice de mercado. As medidas de risco aplicadas são Value at Risk (VaR), Expected Shortfall (ES) and Expectile Value at Risk (EVaR), elas são previstas através do modelo GARCH clássico unido com nove funções de distribuição de probabilidade diferentes e mais por um método não paramétrico. As previsões são avaliadas por funções de perda e backtests de violação. Os resultados indicam que nossa abordagem pode gerar uma função de agregação adequada para processar o risco de um sistema previamente selecionado. / In this work, we deepen the study of systemic risk measurement via aggregation functions. We consider three different portfolios as a proxy for an economic system, these portfolios are consisted in two aggregation functions, based on all U.S. stocks and a market index. The risk measures applied are Value at Risk (VaR), Expected Shortfall (ES) and Expectile Value at Risk (EVaR), they are forecasted via the classical GARCH model along with nine distribution probability functions and also by a nonparametric approach. The forecasts are evaluated by loss functions and violation backtests. Results indicate that our approach can generate an adequate aggregation function to process the risk of a system previously selected.

Neste trabalho, aprofundamos o estudo sobre risco sistêmico via funções de agregação. Consideramos três carteiras diferentes como proxy para um sistema econômico, estas carteiras são consistidas por duas funções de agregação, baseadas em todos as ações do E.U.A, e um índice de mercado. As medidas de risco aplicadas são Value at Risk (VaR), Expected Shortfall (ES) and Expectile Value at Risk (EVaR), elas são previstas através do modelo GARCH clássico unido com nove funções de distribuição de probabilidade diferentes e mais por um método não paramétrico. As previsões são avaliadas por funções de perda e backtests de violação. Os resultados indicam que nossa abordagem pode gerar uma função de agregação adequada para processar o risco de um sistema previamente selecionado. / In this work, we deepen the study of systemic risk measurement via aggregation functions. We consider three different portfolios as a proxy for an economic system, these portfolios are consisted in two aggregation functions, based on all U.S. stocks and a market index. The risk measures applied are Value at Risk (VaR), Expected Shortfall (ES) and Expectile Value at Risk (EVaR), they are forecasted via the classical GARCH model along with nine distribution probability functions and also by a nonparametric approach. The forecasts are evaluated by loss functions and violation backtests. Results indicate that our approach can generate an adequate aggregation function to process the risk of a system previously selected.

Neste trabalho, aprofundamos o estudo sobre risco sistêmico via funções de agregação. Consideramos três carteiras diferentes como proxy para um sistema econômico, estas carteiras são consistidas por duas funções de agregação, baseadas em todos as ações do E.U.A, e um índice de mercado. As medidas de risco aplicadas são Value at Risk (VaR), Expected Shortfall (ES) and Expectile Value at Risk (EVaR), elas são previstas através do modelo GARCH clássico unido com nove funções de distribuição de probabilidade diferentes e mais por um método não paramétrico. As previsões são avaliadas por funções de perda e backtests de violação. Os resultados indicam que nossa abordagem pode gerar uma função de agregação adequada para processar o risco de um sistema previamente selecionado. / In this work, we deepen the study of systemic risk measurement via aggregation functions. We consider three different portfolios as a proxy for an economic system, these portfolios are consisted in two aggregation functions, based on all U.S. stocks and a market index. The risk measures applied are Value at Risk (VaR), Expected Shortfall (ES) and Expectile Value at Risk (EVaR), they are forecasted via the classical GARCH model along with nine distribution probability functions and also by a nonparametric approach. The forecasts are evaluated by loss functions and violation backtests. Results indicate that our approach can generate an adequate aggregation function to process the risk of a system previously selected.

We extend different Description Logics by concrete domains (such as integers and reals) and by aggregation functions over these domains (such as min,max,count,sum), which are usually available in database systems. We present decision procedures for the inference problems satisfiability for these Logics-provided that the concrete domain is not too expressive. An example of such a concrete domain is the set of (nonnegative) integers with comparisons (=,≤, ≤n, ...) and the aggregation functions min, max, count. / This is a new, extended version of a report with the same number. An abridged version has appeared in the Proceedings of the European Conference on Artificial Intelligence, Brighton, UK, 1998.

Integrating sophisticated statistical methods into database management systems is gaining more and more attention in research and industry. One important statistical method is time series forecasting, which is crucial for decision management in many domains. In this context, previous work addressed the processing of ad-hoc and recurring forecast queries. In contrast, we focus on subscription-based forecast queries that arise when an application (subscriber) continuously requires forecast values for further processing. Forecast queries exhibit the unique characteristic that the underlying forecast model is updated with each new actual value and better forecast values might be available. However, (re-)sending new forecast values to the subscriber for every new value is infeasible because this can cause significant overhead at the subscriber side. The subscriber therefore wishes to be notified only when forecast values have changed relevant to the application. In this paper, we reduce the costs of the subscriber by optimizing the notifications sent to the subscriber, i.e., by balancing the number of notifications and the notification length. We introduce a generic cost model to capture arbitrary subscriber cost functions and discuss different optimization approaches that reduce the subscriber costs while ensuring constrained forecast values deviations. Our experimental evaluation on real datasets shows the validity of our approach with low computational costs.

A common optimization technique in data warehouse environments is the use of materialized aggregates. Aggregate processing becomes complex, if partitions of aggregates or queries are materialized and reused later. Most problematic are the implication problems regarding the restriction predicates. We show that in the presence of hierarchies in a multidimensional environment an efficient algorithm can be given to construct - or to derive - an aggregate from one or more overlapping materialized aggregate partitions (set-derivability).