Hasso-Plattner-Institut
Hasso-Plattner-Institut
  
Login
 

Current Projects

The current period is Fall 2018 and started with the HPI Future SOC Lab Day on November 14, 2018. The end is the next HPI Future SOC Lab Day on April 09, 2019. 

35 research projects are using the Lab's IT infrastructure. If you would like to receive more information about one or more projects, please contact us.

Germany

Deep Learning Next-Activity Prediction With Cluster-Based Input Data

Abstract

Predictive process monitoring in the context of adaptive case management is a growing area of research at the intersection of Predictive Analytics and Process Mining. In my masters thesis, I am trying to unify a novel approach in learning preparation through clustering with LSTM neural networks. The results will be compared to those of four papers.

 

Researchers

Principle Investigator: Prof. Dr. Matthias Weske || Contact Author: Felix Wolff || Hasso Plattner Institute, Germany

Scalable Membarriers for Multicore Systems

Abstract

Acknowledged multicasts are a crucial to consistency mechanisms. Sequential propagation is used in practice, resulting in poor performance on larger multicore processors. The impact of tree-based multicast on latency and throughput is evaluated with user-space RCU benchmarks by extending the membarrier systemcall of the Linux kernel.

 

Researchers

Principle Investigator: Prof. Dr.-Ing. Jörg Nolte || Contact Author: Robert Kuban || TU Berlin, Germany

Analysis of cognition in neurophysiological data through deep neural networks

Abstract

Modern neuroimaging provides scientists and medical professionals with a wealth of rich, multi-faceted data – which unfortunately are still analysed with paradigms that have hardly changed in decades. This project proposes a fundamental change by using advances in deep neural networks to find complex, non-linear patterns in neurophysiological data.

 

Researchers

Principle Investigator: Dr. Jeff Hanna || Contact Author: Jeff Hanna || Universitätsklinikum Erlangen, Germany

Neural Networks for ECG Analysis

Abstract

We are using neural networks for finding QRS complexes (representing heart beats) in electrocardiograms (ECGs). QRS detection is traditionally done by means of signal processing which performs poorly on low quality (e.g. noisy) data. We hypothesize that neural networks offer greater accuracy while being more robust to changes in signal quality.

 

Researchers

Principle Investigator: Prof. Dr. Andreas Polze || Contact Author: Jonas Chromik || Hasso Plattner Institute, Germany

Exploring Game-Theoretic Formation of Realistic Networks

Abstract

We have developed an agent-based game-theoretic model which promises a convincing explanation of the structure of real world networks. For investigating the properties of large generated networks and for validating them with real-world data, we apply for the computing power of the Future SOC Lab.

 

Researchers

Principle Investigator: Prof. Dr. Tobias Friedrich || Contact Author: Dr. Pascal Lenzner || Hasso Plattner Institute, Germany

Big Data Analytics for Security

Abstract

In this project which is the continuation of our previous work, we intend to investigate new analytical approaches to derive security value from our data. These approaches include machine learning, graph analytics, and advanced correlation.

 

Researchers

Principle Investigator: Prof. Dr. Christoph Meinel || Contact Author: Dr. Feng Cheng || Hasso Plattner Institute, Germany

Improving Test Suite Generation by Testing Google Play's Top 1000 Apps II

Abstract

Automated generation of test suites tackles the complexity of testing apps. The selection of the generation method is typically done in a trial-and-error fashion due to unknown characteristics of the apps. Analyzing the fitness landscapes of Google Play's top 1000 apps, we want to understand the characteristics and improve the generation method.

 

Researchers

Principle Investigator: Dr. Thomas Vogel || Contact Author: Dr. Thomas Vogel || Humboldt-Universität zu Berlin, Germany

Live Migration Optimization using Machine Learning for Cloud Computing Infrastructure

Abstract

In this project, we study, model and predict live migration overhead using machine learning techniques. It is important to predict the live migration impact on the datacenter performance and to take the migration decision at the optimal times. This is to avoid conflict of interest between the application resources usage and live migration overhead.

 

Researchers

Principle Investigator: Prof. Dr. Christoph Meinel || Contact Author: Mohamed Elsaid || Hasso Plattner Institute, Germany

GPU-Accelerated Causal Inference

Abstract

We want to investigate the use of GPUs for enterprise relevant machine learning algorithms. In particular we aim to adapt an implementation of an algorithm for causal inference to target the GPU as an execution device. This shall reduce the execution time of the algorithm and in addition free CPU cycles to be used by other enterprise applications.

 

Researchers

Principle Investigator: Dr. Matthias Uflacker || Contact Author: Christopher Schmidt || Hasso Plattner Institute, Germany

Measurement-Based Software Performance Engineering for Microservices and Multi-Core Systems

Abstract

DevOps, microservices, and multi-core systems are gaining considerable attraction in research and practice. We would use
the requested HPI Future SOC Lab resources to investigate experimental evaluation of our activities on "DevOps-oriented Load Testing for Microservices" and "Software Performance Engineering for Multi-Core Systems".

 

Researchers

Principle Investigator: Dr.-Ing. Andre van Hoorn || Contact Author: Dr.-Ing. Andre van Hoorn || University of Stuttgart, Germany

A Deeper Look into Bots using Deep Learning

Abstract

Online social networks have become a massive source of information spreading and diffusion. However, users are not the only ones generating content on these platforms. Automated accounts (Bot) have been spreading misinformation and abusing social networks. In this work, we intend to develop a bot detection system using graph convolutional networks.

Researchers

Principle Investigator: Prof. Dr. Christoph Meinel || Contact Author: Ali Alhosseini || Hasso Plattner Institute, Germany

ML-based Classification of Lung Diseases

Abstract

The radiological evaluation of lung tissue is pivotal for an accurate diagnosis of lung diseases. Current advances in machine learning technologies have forwarded the automatic detection of pathological structures in human tissue based on radiographs. Such automated image analysis approaches have great potential to increase the diagnosis precision

 

Researchers

Principle Investigator: Prof. Dr. Oliver Hinz || Contact Author: Benjamin M. Abdel-Karim || Goethe University, Germany

Integrating Hardware Accelerators in Virtualized Environments

Abstract

This project investigates methods for integrating accelerators in distributed, virtualized environments. The project is a continuation from a project started in the spring 2018 period.

 

Researchers

Principle Investigator: Prof. Dr. Andreas Polze || Contact Author: Max Plauth || Hasso Plattner Institute, Germany

Operating Heterogeneous Server System

Abstract

Arbeitslasten wie Maschinelles Lernen, In-Memory Computing sowie ähnlich komplexe Anwendungen stellen vermehrt eine Herausforderung für den Betrieb von Rechenzentren dar. Diese intensive Nutzung von Ressourcen rückt das Energiemanagement im Betrieb in den Fokus.

 

Researchers

Principle Investigator: Prof. Dr. Andreas Polze || Contact Author: Felix Eberhardt || Hasso Plattner Institute, Germany

Using machine learning image analysis to identify abnormal chest X-rays in community-based TB screen

Abstract

In a community-based screening study assessing the health status of a South African population, a prompt detection of lung abnormalities is crucial for tuberculosis diagnosis. Here, we use machine learning to extract abnormal features in chest x-rays from this population study and predict the TB status and other parameters.

 

Researchers

Principle Investigator: Prof. Dr. Christoph Lippert || Contact Author: Jana Fehr || Hasso Plattner Institute, Germany

Body Surface Estimation from Video Data

Abstract

In the context of lower back pain, we want to quantify and analyze motion patterns of the back based on the change of its shape during back-specific clinical examinations. From the video recordings of these, we want to estimate the body surface coordinates by using GPU-based Pose Estimation.

 

Researchers

Principle Investigator: Prof. Erwin Böttinger || Contact Author: Jan Philipp Sachs || Hasso Plattner Institute, Germany

Neuroimaging Biomarker Prediction with Convolutional Neural Networks

Abstract

Here, I propose a project to establish an AI-based method to process brain MRI scans. In this project, a convolutional neural network will be developed to predict structural biomarkers applicable to epidemiological studies and personalized medicine, contributing to the automatic processing of large-scale multimodal medical data.

 

Researchers

Principle Investigator: Prof. Dr. Christoph Lippert || Contact Author: Shahryar Khorasani || Hasso Plattner Institute, Germany

Behavior-based authentication: feature engineering and evaluation based on large user profiles

Abstract

Our project contributes to the field of behavior-based authentication. In the last years we collected a great amount of walking sequences of several people. As this dataset is too large to be processed on a normal machine, we hope to evaluate and improve our authentication model by the support of additional

 

Researchers

Principle Investigator: Prof. Dr. Christoph Meinel || Contact Author: Vera Weidmann || neXenio GmbH, Germany

openhpi-monitoring

Abstract

The openHPI infrastructure yields a lot of monitoring data. This project aims on identifying odd platform behavior and sliding dynamic thresholds using machine learning technology. A system to be created as deliverable of this project seminar should learn from historical data together with known events of system failure or performance issues.

 

Researchers

Principle Investigator: Prof. Dr. Christoph Meinel || Contact Author: Christian Willems || Hasso Plattner Institute, Germany

Europe

WEEVIL- Second (WEEVILS)

Abstract

The WEEVILS project is designed as the extension of the previous one, WEEVIL, which was developed using the RX600S5-1 server from the HPI Future SOC Lab infrastructure. This document details the main aims, scope and schedule development for the WEEVILS project. This project is a continuation of previous one, this submission is a renewal for accessi

 

Researchers

Principle Investigator: Prof. Carlos Juiz || Contact Author: Belen Bermejo || University of the Balearic Islands, Spain

Comparative of techniques for similarity measures applied for Voice Recognition

Abstract

The project is a comparative study between different techniques for Time Series Classification (That are explained in the PDF file) against a dataset of wave audio from kaggle® for Voice Commands for later then make a combination of the best ones to try to obtain a better accuracy than the ones already obtained.

 

Researchers

Principle Investigator: Dr. Leandro dos Santos Coelho || Contact Author: Joao Sauer || Universidade Federal do Paraná, Brazil

Using HPI Resources for Geographical Based Telecommunication Marketing by Visualization of Subscribe

Abstract

We would like to continue a previous project at the Future SOC Lab at HPI on optimization using mobility data. The project is described in the attached PDF file.

 

Researchers

Principle Investigator: Prof. Lars Lundberg || Contact Author: Dr. Julia Sidorova || Blekinge Institute of Technology, Sweden

Quality Assessment of RDF datasets at Large Scale

Abstract

We are working on a novel approach for the quality assessment of large RDF dataset and implement it using an efficient framework for large-scale, distributed and in-memory computations. Within this project, we will focus on performing analysis of the complexity of the computational steps and the data exchange between nodes in the cluster.

 

Researchers

Principle Investigator: Dr. Anisa Rula || Contact Author: Dr. Anisa Rula || University of Milano-Bicocca, Italy

Deep Representation Learning on Large Attributed Graphs

Abstract

Real-world graphs are often associated with a rich set of attributes (e.g. text,image) that can significantly influence interactions, such as in political campaigns. This project wants to develop a Deep Learning unsupervised model able to generate a representation that encodes both the relational structure and node attributes in these large graphs.

 

Researchers

Principle Investigator: Prof. Vincenzina Messina || Contact Author: Dr. Debora Nozza || University of Milano-Bicocca, Italy

Benchmarking FFT on Ethernet Clusters

Abstract

The aim of this project is to complement measurements of Fourier transform on computers with proprietary and expensive interconnects with those using less costly high performance ethernet. Such performance measurements are useful to understand performance tradeoffs for converged big data and high performance computing.

 

Researchers

Principle Investigator: Dr. Benson Muite || Contact Author: Dr. Benson Muite || University of Tartu, Estonia

Blockchains + RFIDs = Verified Supply Chain Management

Abstract

We propose a solution to contrast counterfeiting of luxury fashion items. Each clothing or accessory is associated with a unique and unclonable RFID tag, and at each step of the manufacturing and of the delivery of the items from the production factories to the retail shops the relevant information is stored on a permissioned blockchain.

 

Researchers

Principle Investigator: Prof. Alberto Leporati || Contact Author: Prof. Alberto Leporati || University of Milan-Bicocca, Italy

Text-based Knowledge Graph Embeddings

Abstract

KG is a widespread abstraction model to represent knowledge about entities in a variety of domains, In this proposal we want to use GPU intensive algorithm based on RNN and LSTMs to improve the link prediction ok a KG, moreover we also want to use tree-like structures in a hyperbolic metric space as new embedding approach.

 

Researchers

Principle Investigator: Prof. Andrea Maurino || Contact Author: Prof. Andrea Maurino || University of Milano-Bicocca, Italy

Learning a metapath-embedding

Abstract

As a successor to our previous Future SOC Lab project, we want to investigate how to best learn a distributed representation for meta-paths in large knowledge graphs like Wikidata.

 

Researchers

Principle Investigator: Prof. Dr. Davide Mottin || Contact Author: Sebastian Bischoff || Aarhus University, Denmark

Distributed Computation of Pareto Front for Multi-objective Evolutionary Algorithms

Abstract

Multi-objective Evolutionary Algorithms (MOEAs) have been successfully used for optimising several problems yet when applied to real-world large problems the Pareto Front (PF) construction slows them down maiking infeasible their use in practise. We aim at investigating novel techniques to distribute and parallelise the PF computation.

 

Researchers

Principle Investigator: Associate Prof. Federica Sarro || Contact Author: Associate Professor Federica Sarro || University College London, United Kingdom

A Comprehensive Characterization of HTAP

Abstract

The goal of this project is to understand the HTAP landscape and first develop a benchmark suite that would be representative of the different set of use cases that fall under the HTAP. Then, we are going to numerically compare a subset of available HTAP systems on a large-scale cluster using the proposed benchmark suite.

 

Researchers

Principle Investigator: Associate Professor Pinar Tozun || Contact Author: Associate Professor Pinar Tozun || IT University of Copenhagen, Denmark

City Sensing (continuation of the project)

Abstract

Big mobility and IoT data processing and analytics in urban computing. The planned activities are focused on research and development of methods, tools and software systems for efficient and effective processing, analysis and mining of Big mobility data collected leveraging mobile crowd sensing and IoT paradigms in smart cities.

 

Researchers

Principle Investigator: Prof. Dragan Stojanovic || Contact Author: Prof. Dragan Stojanovic || University of Nis, Serbia

Worldwide

Comparative of techniques for similarity measures applied for Voice Recognition

Abstract

The project is a comparative study between different techniques for Time Series Classification (That are explained in the PDF file) against a dataset of wave audio from kaggle® for Voice Commands for later then make a combination of the best ones to try to obtain a better accuracy than the ones already obtained.

 

Researchers

Principle Investigator: Dr. Leandro dos Santos Coelho || Contact Author: Joao Sauer || Universidade Federal do Paraná, Brazil

ImageNet classification with random depthwise signed convolutional neural networks

Abstract

In recent work we have shown random depthwise convolutional neural networks to attain accuracies better than previous random nets and not far behind in accuracy to trained state of the art networks, especially in the top-k setting. In this proposal we wish to study our network on large image benchmarks such as ImageNet.

 

Researchers

Principle Investigator: Associate Prof. Usman Roshan || Contact Author: Associate Prof. Usman Roshan || New Jersey Institute of Technology, United States of America

Scalable Batch and Real-time Analytics of Trajectory Data

Abstract

We are investigating scalable and polyglot systems for processing trajectory data (relational, doc-oriented, graph-oriented, time-series dbs). Business questions are 2 types OLAP or patterns' mining. We are working on two datasets 1) The Danish maritime data (2TB) and 2) NYC cabs' trips (200GB).

 

Researchers

Principle Investigator: Dr. Rim Moussa || Contact Author: Dr. Rim Moussa || National Engineering School of Carthage, Tunisia

On-Demand Representation Learning for Heterogeneous Data

Abstract

We propose an On-Demand Representation Learning service that flexibly learns suitable representations for items submitted by a client. It draws on a combination of heterogeneous massive-scale sources, thereby enabling the client's machine learning algorithms to better cope with new signals that never occurred in their domain-specific training data.

 

Researchers

Principle Investigator: Prof. Dr.-Ing. Gerard de Melo || Contact Author: Prof. Dr.-Ing. Gerard de Melo || Rutgers University, United States of America

Modelling of aluminum reduction processes with machine learning approaches and mathematical models

Abstract

Metallurgy processes are multidisciplinary and very hard to model and control, however recent advances in technology have helped many industries to overcome such challenges. This project exploits the use of in-memory processing, machine learning algorithms, as well as widely used mathematical models to make predictions and detect future failures.

 

Researchers

Principle Investigator: Prof. Dr. Roberto Celio Limao de Oliveira || Contact Author:  Fabio Mendes Soares || Federal University of Para, Brazil