arXiv subjects

Dimitrios Tsoumakos

Publications and source records attributed to Dimitrios Tsoumakos.

8 recordsLinked to original sources

arXiv2025

Analytics Modelling over Multiple Datasets using Vector Embeddings

The massive increase in the data volume and dataset availability for analysts compels researchers to focus on data content and select high-quality datasets to enhance the performance of analytics operators. While selecting high-quality data significantly boosts analytical accuracy and efficiency, the exact process is very challenging given large-scale dataset availability. To address this issue, we propose a novel methodology that infers the outcome of analytics operators by creating a model from the available datasets. Each dataset is transformed to a vector embedding representation generated by our proposed deep learning model NumTabData2Vec, where similarity search are employed. Through experimental evaluation, we compare the prediction performance and the execution time of our framework to another state-of-the-art modelling operator framework, illustrating that our approach predicts analytics outcomes accurately, and increases speedup. Furthermore, our vectorization model can project different real-world scenarios to a lower vector embedding representation accurately and distinguish them.

cs.LG↗

arXiv2025

Chunked Data Shapley: A Scalable Dataset Quality Assessment for Machine Learning

As the volume and diversity of available datasets continue to increase, assessing data quality has become crucial for reliable and efficient Machine Learning analytics. A modern, game-theoretic approach for evaluating data quality is the notion of Data Shapley which quantifies the value of individual data points within a dataset. State-of-the-art methods to scale the NP-hard Shapley computation also face severe challenges when applied to large-scale datasets, limiting their practical use. In this work, we present a Data Shapley approach to identify a dataset's high-quality data tuples, Chunked Data Shapley (C-DaSh). C-DaSh scalably divides the dataset into manageable chunks and estimates the contribution of each chunk using optimized subset selection and single-iteration stochastic gradient descent. This approach drastically reduces computation time while preserving high quality results. We empirically benchmark our method on diverse real-world classification and regression tasks, demonstrating that C-DaSh outperforms existing Shapley approximations in both computational efficiency (achieving speedups between 80x - 2300x) and accuracy in detecting low-quality data regions. Our method enables practical measurement of dataset quality on large tabular datasets, supporting both classification and regression pipelines.

cs.LG↗

arXiv2018

Graph Operator Modeling over Large Graph Datasets

As graph representations of data emerge in multiple domains, data analysts need to be able to intelligently select among a magnitude of different data graphs based on the effects different graph operators have on them. Exhaustive execution of an operator over the bulk of available data sources is impractical due to the massive resources it requires. Additionally, the same process would have to be re-implemented whenever a different operator is considered. To address this challenge, this work proposes an efficient graph operator modeling methodology. Our novel approach focuses on the inputs themselves, utilizing graph similarity to infer knowledge about input graphs. The modeled operator is only executed for a small subset of the available graphs and its behavior is approximated for the rest of the graphs using machine learning techniques. Our method is operator-agnostic, as the same similarity information can be reused for modeling multiple graph operators. We also propose a family of similarity measures based on the degree distribution that prove capable of producing high quality estimations, comparable or even surpassing other much more costly, state-of-the-art similarity measures. Our evaluation over both real-world and synthetic graphs indicates that our method achieves extremely accurate modeling of many commonly encountered operators, managing massive speedups over a brute-force alternative.

cs.SI↗

arXiv2017

Skyline Queries in O(1) time?

The skyline of a set $P$ of points ($SKY(P)$) consists of the "best" points with respect to minimization or maximization of the attribute values. A point $p$ dominates another point $q$ if $p$ is as good as $q$ in all dimensions and it is strictly better than $q$ in at least one dimension. In this work, we focus on the static $2$-d space and provide expected performance guarantees for $3$-sided Range Skyline Queries on the Grid, where $N$ is the cardinality of $P$, $B$ the size of a disk block, and $R$ the capacity of main memory. We present the MLR-tree, which offers optimal expected cost for finding planar skyline points in a $3$-sided query rectangle, $q=[a,b]\times(-\infty,d]$, in both RAM and I/O model on the grid $[1,M]\times [1,M]$, by single scanning only the points contained in $SKY(P)$. In particular, it supports skyline queries in a $3$-sided range in $O(t\cdot t_{PAM}(N))$ time ($O((t/B)\cdot t_{PAM}(N))$ I/Os), where $t$ is the answer size and $t_{PAM}(N)$ the time required for answering predecessor queries for $d$ in a PAM (Predecessor Access Method) structure, which is a special component of MLR-tree and stores efficiently root-to-leaf paths or sub-paths. By choosing PAM structures with $O(1)$ expected time for predecessor queries under discrete $μ$-random distributions of the $x$ and $y$ coordinates, MLR-tree supports skyline queries in optimal $O(t)$ expected time ($O(t/B)$ expected number of I/Os) with high probability. The space cost becomes superlinear and can be reduced to linear for many special practical cases. If we choose a PAM structure with $O(1)$ amortized time for batched predecessor queries (under no assumption on distributions of the $x$ and $y$ coordinates), MLR-tree supports batched skyline queries in optimal $O(t)$ amortized time, however the space becomes exponential. In dynamic case, the update time complexity is affected by a $O(log^{2}N)$ factor.

cs.DB↗

arXiv2017

A Decision Tree Based Approach Towards Adaptive Profiling of Distributed Applications

The adoption of the distributed paradigm has allowed applications to increase their scalability, robustness and fault tolerance, but it has also complicated their structure, leading to an exponential growth of the applications' configuration space and increased difficulty in predicting their performance. In this work, we describe a novel, automated profiling methodology that makes no assumptions on application structure. Our approach utilizes oblique Decision Trees in order to recursively partition an application's configuration space in disjoint regions, choose a set of representative samples from each subregion according to a defined policy and return a model for the entire space as a composition of linear models over each subregion. An extensive evaluation over real-life applications and synthetic performance functions showcases that our scheme outperforms other state-of-the-art profiling methodologies. It particularly excels at reflecting abnormalities and discontinuities of the performance function, allowing the user to influence the sampling policy based on the modeling accuracy and the space coverage.

cs.DC↗

arXiv2014

Cloud elasticity using probabilistic model checking

Cloud computing has become the leading paradigm for deploying large-scale infrastructures and running big data applications, due to its capacity of achieving economies of scale. In this work, we focus on one of the most prominent advantages of cloud computing, namely the on-demand resource provisioning, which is commonly referred to as elasticity. Although a lot of effort has been invested in developing systems and mechanisms that enable elasticity, the elasticity decision policies tend to be designed without guaranteeing or quantifying the quality of their operation. This work aims to make the development of elasticity policies more formalized and dependable. We make two distinct contributions. First, we propose an extensible approach to enforcing elasticity through the dynamic instantiation and online quantitative verification of Markov Decision Processes (MDP) using probabilistic model checking. Second, we propose concrete elasticity models and related elasticity policies. We evaluate our decision policies using both real and synthetic datasets in clusters of NoSQL databases. According to the experimental results, our approach improves upon the state-of-the-art in significantly increasing user-defined utility values and decreasing user-defined threshold violations.

cs.DC↗

arXiv2014

Rapid AkNN Query Processing for Fast Classification of Multidimensional Data in the Cloud

A $k$-nearest neighbor ($k$NN) query determines the $k$ nearest points, using distance metrics, from a specific location. An all $k$-nearest neighbor (A$k$NN) query constitutes a variation of a $k$NN query and retrieves the $k$ nearest points for each point inside a database. Their main usage resonates in spatial databases and they consist the backbone of many location-based applications and not only (i.e. $k$NN joins in databases, classification in data mining). So, it is very crucial to develop methods that answer them efficiently. In this work, we propose a novel method for classifying multidimensional data using an A$k$NN algorithm in the MapReduce framework. Our approach exploits space decomposition techniques for processing the classification procedure in a parallel and distributed manner. To our knowledge, we are the first to study the classification of multidimensional objects under this perspective. Through an extensive experimental evaluation we prove that our solution is efficient and scalable in processing the given queries. We investigate many different perspectives that can affect the total computational cost, such as different dataset distributions, number of dimensions, growth of $k$ value and granularity of space decomposition and prove that our system is efficient, robust and scalable.

cs.DB↗

arXiv2012

SART: Speeding up Query Processing in Sensor Networks with an Autonomous Range Tree Structure

We consider the problem of constructing efficient P2P overlays for sensornets providing "Energy-Level Application and Services". The method presented in \cite{SOPXM09} presents a novel P2P overlay for Energy Level discovery in a sensornet. However, this solution is not dynamic, since requires periodical restructuring. In particular, it is not able to support neither join of sensor\_nodes with energy level out of the ranges supported by the existing p2p overlay nor leave of \emph{empty} overlay\_peers to which no sensor\_nodes are currently associated. On this purpose and based on the efficient P2P method presented in \cite{SPSTMT10}, we design a dynamic P2P overlay for Energy Level discovery in a sensornet, the so-called SART (Sensors' Autonomous Range Tree). The adaptation of the P2P index presented in \cite{SPSTMT10} guarantees the best-known dynamic query performance of the above operation. We experimentally verify this performance, via the D-P2P-Sim simulator (D-P2P-Sim is publicly available at http://code.google.com/p/d-p2p-sim/).

cs.DC↗