Catedra de Calculatoare Facultatea de Automatica si Calculatoare Universitatea Politehnica Bucuresti, Romania

Address Splaiul Independentei 313a Bucuresti Romania

Burse de Excelenta in Cercetare - Competitia 2019

Scopul acestor burse este de a oferi cercetatorilor cu potential, aflati la inceputul carierei, sprijin financiar pe o perioada de 2-3 ani, pentru realizarea de cercetare independenta si cu impact semnificativ.

Criterii de evaluare: un singur criteriu, si anume excelenta. Fiecare propunere va fi evaluata de comitetul organizator din Departamentul de Calculatoare, pentru triaj. Cele mai promitatoare propuneri vor fi trimise catre experti evaluatori externi; fiecare propunere finantata va avea minim o recenzie din partea unui expert international recunoscut in domeniu.

Cuantumul finantarii: 1000 eur/luna.

Durata proiectelor: intre 24 si 36 luni (durata trebuie sa reflecte complexitatea proiectului propus)

Domenii cercetare: toate directiile de cercetare din computer science incluzind (dar nu numai):

• Sisteme: sisteme de operare, retele de calculatoare, compilatoare, baze de date.
• Arhitecturi hardware: arhitecturi paralele de calcul, High Performance Computing, etc.
• Teorie: verificare, metode numerice, algoritmica. etc.
• Securitate
• Machine learning, Inteligenta artificiala, Computer vision
• Algoritmi si sisteme de Loocalizare
• Virtual and augmented reality, grafica
• Robotica
• Alte domenii relevante pentru Computer Science.

Aplicabilitatea programului:

• Cadre didactice aflate la inceput de cariera din Facultatea de Automatica si Calculatoare, Departamentul de Calculatoare
• Candidati externi (din afara catedrei, fie din Romania sau din afara ei) care au finalizat studiile doctorale si doresc sa se alature Departamentului de Calculatoare.
• Candidatii au obtinut titlul de doctor dupa data de 1 martie 2014.
• Aplicantul nu a fost directorul unui proiect de cercetare competitiv national sau international (se exclud proiectele GeX si POSDRU).
• Aplicantul se angajeaza sa dedice minim 50% din timpul sau proiectului de cercetare propus (cu un minim de 80h pe luna).
• Fiecare aplicant poate depune o singura propunere de proiect, fie individual sau intr-un colectiv de cercetare.
• Incurajam aplicatiile individuale, insa in cazuri exceptionale vor fi considerate pentru finantare si aplicatiile elaborate in comun de doi aplicanti; in acest caz, fiecare aplicant trebuie sa indeplineasca criteriile de eligibilitate, propunerea de proiect trebuie sa aiba o complexitate sporita, si sa demonstreze necesitatea colaborarii pentru succesul cercetarii.
• Aplicantul nu a primit un Research Grant in competitia 2018.

Proiectul ideal trebuie sa articuleze clar problema abordata si importanta ei, solutia propusa, si sa descrie de ce aplicantul (sau echipa) este potrivita pentru a duce la indeplinire acest proiect. Vor fi preferate proiectele indraznete de tip "high risk/high gain" celor cu caracter incremental.

Rezultate dorite: scopul burselor este sa incurajeze cercetarea independenta si de impact. Pentru publicatii se urmareste calitatea in detrimentul cantitatii: articole in conferinte A si A* (si workshop-uri asociate acestora) si jurnale din zona rosie/galbena. Articolele in conferinte/jurnale lipsite de vizibilitate internationala si prestigiu sunt descurajate.

Evaluarea proiectelor in timpul derularii: anual, cercetatorul (sau echipa) care a castigat un grant de cercetare va preda un document tehnic (e.g. raport tehnic sau articol) comitetului de evaluare ce sintetizeaza munca din acel an (prototip software, rezultate experimentale, teoreme, etc). Cercetatorul va sustine deasemenea o prezentare in fata comitetului de evaluare. In urma evaluarii documentului si a prezentarii, comitetul va decide continuarea sau intreruperea bursei in functie de calitatea muncii depuse si a angajamentului cercetatorului.

Termen limita de depunere a propunerilor: 15 Martie 2019 (23:59 PST).

Depunerea propunerilor se poate efectua online la aceasta adresa.

Formatul propunerii: numarul total de pagini este de maxim 12 excluzand referintele. Dimensiunea fontului folosit trebuie sa fie minim de 10 pt, cu single spacing intre linii (12pt lead). Dimensiunea maxima a blocului de text este de 16.5cm latime si 23cm inaltime. Propunerea trebuie sa fie elaborata in limba engleza si sa includa urmatoarele sectiuni:

1. Abstract
2. Problem statement
3. State of the art overview
4. Proposed approach and feasibility
5. The PI: brief curriculum vitae.
6. Track record: top 5 papers and description of how prior work relates to the proposed project.

Sursa de finantare: resurse interne CRC

Comitet organizator:

• Costin Raiciu (sisteme, retele, securitate)
• Emil Slusanschi (arhitecturi de calculatoare, high performance computing)
• Dragos Niculescu (retele wireless, probleme de localizare)
• Lorina Negreanu (teorie, algoritmi, verificare)
• Marius Leordeanu(computer vision, machine learning, inteligenta artificiala)
• Dan Tudose (sisteme embedded, Internet of Things)
• Razvan Rughinis (retele, aplicabilitate in industrie)

Contact: Costin Raiciu (costin.raiciu@cs.pub.ro) pentru clarificari si probleme ale site-ului pentru aplicatii.

Rezultate Competitia 2019

Au fost depuse 7 proiecte, din care au fost selectate doua proiecte pentru finantare pentru o perioada de 3 ani:

1. Principal Investigator Cristian Tranca
   
   Proiect (finantat de Cegeka Romania) In-line Embedded Industrial Firewall
   
   Abstract
   In the context of the so-called Industry 4.0, the integration of Internet Technology into industrial facilities raises security issues that were practically non-existent until now. The rapid evolution of remote-control solutions via Internet, corroborated with the necessity of backward compatibility and real-time predictable communications requirements, specific for industrial protocols, can generate security threats with tragic consequences. Industrial communication in many industrial networks is done using either serial bus line - drop-down topology, either Ethernet. Industrial equipment is designed to operate for tens of years without intervention, making disruptive changes and upgrades in industrial communication networks hard to implement. We propose to tackle industrial security concerts through solutions provided near the protected device, to decentralize the rule-based filtering and encryption (where necessary) and to design low-power, energy efficient embedded in-line industrial firewalls. The benefits will be threefold: (1) rule-based packet filtering will be done directly at the destination (slave) device or in a direct network node (for serial bus lines) thus preventing malicious commands from inside the network reaching the target device; (2) packet analysis will be done on the fly, simple and energy efficient, by temporarily jamming the network or breaking the network line (where possible); (3) the inline plug-and-play architecture will allow inserting/ removing the firewall device without any need of reconfiguration of any other networked equipment.
   
   
2. Principal Investigator Daniel Rosner
   
   Proiect (finantat de UPB)
   HEIMDAL NON-INTRUSIVE LOAD MONITORING SYSTEM A BOTTOM UP APPROACH FOR A GROWING NEED
   
   Abstract
   We are in a golden age of machine learning (ML), and its impact is wide-reaching: from the way in which authorities make infrastructure decisions, to how companies browse business opportunities, to techniques for diagnosis or medication prescription, all are being significantly augmented by ML. One specific area where there is allot of hope from ML-based solutions is the Non-Intrusive Load Monitoring (NILM) systems area – devices that can accurately identify the active consumers in a house, company office or industrial facility. The challenge is to minimize the number of such measuring devices. Additionally, for the approach to be impactful, the algorithms should be robust enough to generate meaningful information even when collection takes place on real world physical systems, outside of controlled (experimental /test set-ups) environments. Such a system, if truly efficacious in real world scenarios, can provide valuable insights to be leveraged by:
   • Home and small & medium businesses, to optimize their cost and increase energy savings;
   • Energy companies to optimize their electric power provisioning, offer custom plans for their clients, as well as optimize other business actions based on data driven decisions.
   The current proposal aims to start off a new base for building NILM solutions that could reach the maturity level required to impact the energy sector. As such, we propose to:
   1. improve the quality of available data for building ML algorithms for NILM applications;
   2. determine an optimal balance between sampling rate and accuracy, in order to lay the foundation for a NILM hardware systems design guideline, while also considering practical processing power, memory limitations and acceptable data transfer and storage rates;
   3. introduce a set of signal pre-processing, signal identification and ML-based labeling and identification algorithms that can deliver a high accuracy rate in real world scenarios.
   To achieve this, we will follow a bottom-up approach:
   1. construct the physical infrastructure for building up a high resolution, high variance data set, with both central measuring devices, as well as “witness” (“observer”) secondary measuring devices;
   2. enrich the data set; test different approaches and variants for sample rates, pre- processing, data compression and transmission, signal identification, feature identification, data storage;
   3. tally the available ML algorithmic approaches, evaluate their suitability on a number of technical-economic indicators and perform preliminary training;
   4. construct an ML algorithm based on consensus between at least two basic ML approaches (one, primary and the other, secondary), optimize it in terms of required processing power and data transfer rates and quantities and showcase it on the central monitoring system;
   5. (later stage) the above data and technological infrastructure, when deployed in real world scenarios, will open other research questions in terms of security, privacy, energy efficiency; these will be considered in a later project.
   When complete, this research will lay the foundation for building reliable NILM systems, from physical sensors to data-based identification.