Research Projects

The dynamics of social networks, namely the way connections are created and strengthened between individuals, is a very complicated process with many factors interfering in the emergence and evolution of social ties. Current research tries to better explain the high complexity of such network structures, by proposing corresponding algorithmic models for the evolution of topologies and influence propagation. Nevertheless, state of the art models have multiple limitations, as they typically make simplifying assumptions about opinion interaction mechanisms with fixed thresholds, non-dynamic topologies, or pervasive opinion sources.

This project comes to push the boundaries of scientific understanding forward, on several levels, by placing novel and existing pieces of the puzzle together, in terms of better predicting the spread of opinion over large social temporal networks. The main objectives of this project are:

• To define an original temporal agent-based interaction model (with a dynamic time-aware threshold);
• To explore and define, through mathematical modeling and computer simulation, novel trade-off strategies for improving opinion diffusion coverage, while maintaining a minimal cost of operation for engaged spreader agents;
• To improve the prediction accuracy of opinion distribution by integrating the temporal attenuation paradigm, with direct applicability in electoral poll forecasting;
• To combine the obtained interaction model, with diffusion strategies, and temporal poll prediction into a simulator application for defining a robust opinion poll prediction framework.

The motivation of these research goals is supported by the social and economic impact potential of the project. Namely, inferring the underlying dynamics of social interaction is of outstanding present interest, since it has direct applicability in viral marketing, political science, and even epidemiology, for predicting the spread of a commercial, a rumor, or a virus.

The aim of the project is to increase the research and innovation capacity of UPT by creating a high-performance, heterogenous, energy efficient, private cloud computing node based on open technologies, which is connected to national and international networks of research cloud infrastructures, and which has direct applicability in the collection, storage, analysis, distribution and the protection of the massive data produced by the research and innovation initiatives carried out in the western region of Romania.

The main target group of the project consists of all the UPT researchers and PhD students, regardless of the research field in which they operate, who can benefit from the services provided by IT-specific tools, provided in the form of centralized computing services, storage and productivity. The project also targets public and private research partners which collaborate with the UPT at local, national and international level.

Objectives:

• Implementation of an energy-efficient, HPC-type of computing and data storage infrastructure to provide flexible to manage and resource scalable virtual environments
• Implementation of a large capacity data collection, storage and interchange platform, to be provided as a service (PaaS - Platform as a Service). Three categories of specific data are considered for validating this implementation: (a) crowdsourcing/ IoT type of data, (b) multimedia type of data and (c) geographical type of data
• Implementation of a platform for cloud-based functions (FaaS - Function as a Service), for data processing and analysis. The project also includes the implementation of a demo deep learning application for the intelligent selection of a subset of images (at least 1M images), based on diversity estimation
• Implementation of a local solution for managing and storing of the personal data which results from the research and innovation activities within UPT, according to the new GDPR standards

An overview of the hardware implementation results, from the perspective of the nodes assembly and their capabilities, including the storage and FPGA nodes:

• Total of processing and storage nodes: 19
• Total of processing and storage cores: 352
• Total memory capacity: 4000 GB
• Total storage capacity; 598 TB (raw), 417 TB (usable)
• Node connectivity: 40 Gbps
• GPU boards: 4 x NVIDIA Tesla T4 16GB, Passive, Single Wide, Full Height GPU
• FPGA modules:
  • 2 x Intel FPGA PAC with Arria10 GX, 70W FH
  • 2 x Intel PAC D5005, Stratix10, 32GB, 215W, Double Wide, Full Height, QSFP28 FPGA

In a rapidly evolving world, with growing population and accelerated access to online media, the need to understand the structures and behavior of human society has become more important than ever. Having emerged as an interdisciplinary field in the 21st century, social networks analysis is on the quest to understand opinion formation and diffusion from a scientific point of view. The formation of social networks, namely the way connections are created and strengthened, is a very complex process, and there are many factors that interfere for the emergence and evolution of social ties. Notable research efforts try to explain the high complexity of such network structures, by proposing corresponding algorithmic models. Nevertheless, available models have many limitations as they typically assume opinion interaction mechanisms based on fixed thresholds, static topologies, or omnipresent opinion sources.

This project comes to improve our understanding of opinion diffusion in emergent social networks. Consequently, to build models that are aware of these phenomena, I propose fundamental topological analysis of empirical data - using network motifs, community detection algorithms and statistics - to understand the behavioral patterns and centralities which have an impact on spatial and temporal distribution of opinion. As opposed to most existing opinion interaction models, I propose a temporal opinion injection model which evolves over time according to basic human traits and underlying social topology. Also, by employing discrete event simulation on real-time gathered social network data, I propose the implementation of an online platform to offer improved prediction on poll outcomes for socially-relevant topics.

In the wake of big data analytics, this project sets out to push the boundaries of scientific understanding of opinion dynamics in social networks by analyzing how the underlying network topology influences communication patterns and polarization of opinion.

A software tool (desktop and mobile) has been developed for the efficiency assessment of hydraulic generators' operation installed in Timisoara's drinkable water system and beyond. The diagnostic is computed based on QR codes and the traffic light system using three colors: green, for normal operation, yellow, for dangerous operation which requires a planned stop and finally red, for abnormal operation requiring an emergency halt.

The diagnostic system has been validated using laboratory experiments and several selected pumps in service for daylight and nighttime use at AQUATIM. Timisoara has more than 300.000 inhabitants, 30.000 foreign students and several hundreds km of pipes delivering drinkable water to the citizens.

The Principal Investigator of the project was interviewed by TeleU about the project results in this video clip:

The objective of this project is to extend the services offered to the clients by the Piconet company, which is the national leader of city surface parking management systems. The aim is to develop a robust method for monitoring the parking occupancy based on processing of images captured by surveillance cameras. This method has to adapt to harsh weather conditions and to changing in illumination due to some natural causes like clouds or artificial ones like public lighting. It also has to adapt to a large variety of cameras’ deployment angles, and to learn environment changes. It has to offer a good support for future extension of services – e.g. automatic car plates recognition or automatic payment methods.

The company plan to use the method developed here to implement a mobile application that can offer an overview of parking occupancy to the clients for an entire area managed by the company. The utility of this solution is to save the client time spent in finding a parking place, especially in crowded central city areas, and it was already requested by many existing clients. Moreover, it will ensure some other benefits like saving the fuel consumption and reducing the pollution generated by cars that creates overhead traffic on the way to find a free parking place. In order to ensure the desired accuracy this method will combine image processing algorithms with existing statistical information collected by the company, and with learned data. In addition it should include a simple and clear procedure for further system deployment.

The result of research will be a functional prototype that can be integrated in the existing company parking management system.

Recent research indicates Chronic Obstructive Pulmonary Disease (COPD) as the third cause of death and one of the main impediments to the quality of living in today’s society. COPD is defined as the clinical condition which reduces pulmonary capacity; it is not reversible, however if diagnosed at an early phase, its evolution can be controlled. Unfortunately, the early detection of COPD is a difficult task, and often time people are diagnosed when they are already in an advanced stage.

Capitalizing on recent research results which indicate sensor systems, mobile, and Internet of Things solutions as very useful for monitoring and managing COPD, we propose a personal, integrated prototype system for early detection and evolution prediction of COPD. As such, we intend to build a sensor network that gathers multiple physiological signals and a mobile application that extracts the multi-fractal spectra as mere signatures of these signals. Then, the mobile system will integrate the physiologic signatures with anthropometric and other individual clinical data. On the server side, we will collect the integrated data from a population of individuals, to build a complex network model of patients. Indeed, recent papers indicate the complex network model as very useful for generating COPD predictions. To this end, we will employ modularity clustering and network layout tools to build prediction models for both early detection and evolution prediction of COPD. The prediction model will be instantiated as a smartphone application and tested in order to assess its predictive capacity.

In order to undertake the objectives of our proposal, we assemble a multidisciplinary team, consisting of computer engineers (hardware and software) and specialized medical doctors. The engineering teams will build the hardware and software parts of our demonstration model, whereas the medical team will provide the necessary medical expertise, in order to test and validate the engineering model.

Wireless networks of sensors and smart devices (WSN) are an extremely interesting topic, at the confluence of engineering fields with enormous impact on worldwide society: digital networks, wireless communications, and miniature embedded digital devices.

Aware of the severe requirements and challenges raised by current applications in this area, we propose a new paradigm - Time and Energy Efficiency (T: or TEE). The goal of the project is to develop an integrated real-time and energy efficient inter-operation framework for networks of smart sensors and devices - TEEFIOS.

The main proposed objectives focus on three distinct layers: (a) T:Node, a hardware-software environment and methodology for designing and assessing real-time behavior and efficient energy consumption of embedded devices, (b) T:YNet, a system for the development and analysis of TEE communication in wireless ad-hoc networks, and (c) T:PIlot, a methodology for the power management of the entire network. An integrated set of tools, benchmarks and databases will also be created to help advanced developers and researchers in the WSN area apply the TEE paradigm to applications with high impact. All the results will be validated using real-life case studies.

Existing resources include a postdoc and PhD students with direct interest in these fields, balanced by senior researchers with high expertise and international visibility, all supported by a well equipped environment - the DSPLabs.

Project objectives:

• Increasing educational exchanges through a common cross-border approach in the area of technical education by implementing an Education web-cast system;
• Creating a cross-border partnership between the Faculty of Automation and Computers from Timisoara and the Technical Faculty from Zrenjanin;
• Improving the quality of education for the students and pupils from the border area Increasing the overall competitiveness of the economy in the border area.

Project results: EduWebCast systems, lectures that will be broadcasted through the portal, users of the portal, meetings attended by members of both teams, information and publicity events, conferences.

Target groups: University students from the two universities, High school pupils in the cross border area, Graduated students that are employed especially in companies that are located in the border region.

University of Debrecen, as main partner, and the Politehnica University of Timisoara, as cross-border partner, submit a proposal with the stated goal of integrating in a cross-border communication system the local network systems of the two partners. The overall objective of the project is to enable enhanced capacity for cross-country cooperation and interaction between and within the participating universities by providing high quality WiFi system and IP streaming system for the students, professors and researchers at the University of Debrecen and Politehnica University of Timisoara, aiming at supporting the synchronization of educational, research and development, and other scientific activities of the cooperating universities.

This project influences the cooperation activity of the universities involved, facilitating collaboration between the target groups, university students, researchers.

Personal computing currently faces a rapid trend from desktop machines towards mobile services, accessed via tablets, smartphones and similar terminal devices. With respect to computing power, today´s handheld devices are similar to Cray-2 supercomputers from the 1980s. Due to higher computational load (e.g. via multimedia apps) and the variety of radio interfaces (such as WiFi, 3G, and LTE), modern terminals are getting increasingly energy hungry. For instance, a single UMTS upload or a video recording process on today´s smartphones may consume as much as 1.5 Watts, i.e. roughly 50% of the maximal device power. In the near future, higher data rates and traffic, advanced media codecs, and graphics applications will ask for even more energy than the battery can deliver. At the same time, the power density limit might lead to a significant share of “Dark Silicon” at 22nm CMOS and below. Obviously, disruptive energy optimizations are required that go well beyond traditional technologies like DVFS (dynamic voltage and frequency scaling) and power-down of temporarily unused components.

The GEMSCLAIM project aims at introducing novel approaches for reducing this “greed for energy”, thereby improving the user experience and enabling new opportunities for mobile computing. The focus is on three novel approaches: (1) cross layer energy optimization, ranging from the compiler over the operating system down to the target HW platform, (2) efficient programming support for energy-optimized heterogeneous Multicore platforms based on energy-aware service level agreements (SLAs) and energy-sensitive tunable parameters, and (3) introducing energy awareness into Virtual Platforms for the purpose of dynamically customizing the HW architecture for energy optimization and online energy monitoring and accounting. GEMSCLAIM will provide new methodologies and tools in these domains and will quantify the potential energy savings via benchmarks and a HW platform prototype.