Computational inference and control of quality in multimedia services

Research output: ThesisPhd Thesis 1 (Research TU/e / Graduation TU/e)

452 Downloads (Pure)

Abstract

Quality is the degree of excellence we expect of a service or a product. It is also one of the key factors that determine its value. For multimedia services, understanding the experienced quality means understanding how the delivered delity, precision and reliability correspond to the users' expectations. Yet the quality of multimedia services is inextricably linked to the underlying technology. It is developments in video recording, compression and transport as well as display technologies that enables high quality multimedia services to become ubiquitous. The constant evolution of these technologies delivers a steady increase in performance, but also a growing level of complexity. As new technologies stack on top of each other the interactions between them and their components become more intricate and obscure. In this environment optimizing the delivered quality of multimedia services becomes increasingly challenging. The factors that aect the experienced quality, or Quality of Experience (QoE), tend to have complex non-linear relationships. The subjectively perceived QoE is hard to measure directly and continuously evolves with the user's expectations. Faced with the diculty of designing an expert system for QoE management that relies on painstaking measurements and intricate heuristics, we turn to an approach based on learning or inference. The set of solutions presented in this work rely on computational intelligence techniques that do inference over the large set of signals coming from the system to deliver QoE models based on user feedback. We furthermore present solutions for inference of optimized control in systems with no guarantees for resource availability. This approach oers the opportunity to be more accurate in assessing the perceived quality, to incorporate more factors and to adapt as technology and user expectations evolve. In a similar fashion, the inferred control strategies can uncover more intricate patterns coming from the sensors and therefore implement farther-reaching decisions. Similarly to natural systems, this continuous adaptation and learning makes these systems more robust to perturbations in the environment, longer lasting accuracy and higher eciency in dealing with increased complexity. Overcoming this increasing complexity and diversity is crucial for addressing the challenges of future multimedia system. Through experiments and simulations this work demonstrates that adopting an approach of learning can improve the sub jective and objective QoE estimation, enable the implementation of ecient and scalable QoE management as well as ecient control mechanisms.
Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
  • Department of Electrical Engineering
Supervisors/Advisors
  • Liotta, Antonio, Promotor
Award date5 Mar 2013
Place of PublicationEindhoven
Publisher
Print ISBNs978-90-386-3355-8
DOIs
Publication statusPublished - 2013

Fingerprint

Multimedia services
Video recording
Multimedia systems
Expert systems
Artificial intelligence
Display devices
Availability
Feedback
Sensors
Experiments

Cite this

Menkovski, V.. / Computational inference and control of quality in multimedia services. Eindhoven : Technische Universiteit Eindhoven, 2013. 160 p.
@phdthesis{68ed9a9c6e68413a85cefd8b18a658de,
title = "Computational inference and control of quality in multimedia services",
abstract = "Quality is the degree of excellence we expect of a service or a product. It is also one of the key factors that determine its value. For multimedia services, understanding the experienced quality means understanding how the delivered delity, precision and reliability correspond to the users' expectations. Yet the quality of multimedia services is inextricably linked to the underlying technology. It is developments in video recording, compression and transport as well as display technologies that enables high quality multimedia services to become ubiquitous. The constant evolution of these technologies delivers a steady increase in performance, but also a growing level of complexity. As new technologies stack on top of each other the interactions between them and their components become more intricate and obscure. In this environment optimizing the delivered quality of multimedia services becomes increasingly challenging. The factors that aect the experienced quality, or Quality of Experience (QoE), tend to have complex non-linear relationships. The subjectively perceived QoE is hard to measure directly and continuously evolves with the user's expectations. Faced with the diculty of designing an expert system for QoE management that relies on painstaking measurements and intricate heuristics, we turn to an approach based on learning or inference. The set of solutions presented in this work rely on computational intelligence techniques that do inference over the large set of signals coming from the system to deliver QoE models based on user feedback. We furthermore present solutions for inference of optimized control in systems with no guarantees for resource availability. This approach oers the opportunity to be more accurate in assessing the perceived quality, to incorporate more factors and to adapt as technology and user expectations evolve. In a similar fashion, the inferred control strategies can uncover more intricate patterns coming from the sensors and therefore implement farther-reaching decisions. Similarly to natural systems, this continuous adaptation and learning makes these systems more robust to perturbations in the environment, longer lasting accuracy and higher eciency in dealing with increased complexity. Overcoming this increasing complexity and diversity is crucial for addressing the challenges of future multimedia system. Through experiments and simulations this work demonstrates that adopting an approach of learning can improve the sub jective and objective QoE estimation, enable the implementation of ecient and scalable QoE management as well as ecient control mechanisms.",
author = "V. Menkovski",
year = "2013",
doi = "10.6100/IR751521",
language = "English",
isbn = "978-90-386-3355-8",
publisher = "Technische Universiteit Eindhoven",
school = "Department of Electrical Engineering",

}

Menkovski, V 2013, 'Computational inference and control of quality in multimedia services', Doctor of Philosophy, Department of Electrical Engineering, Eindhoven. https://doi.org/10.6100/IR751521

Computational inference and control of quality in multimedia services. / Menkovski, V.

Eindhoven : Technische Universiteit Eindhoven, 2013. 160 p.

Research output: ThesisPhd Thesis 1 (Research TU/e / Graduation TU/e)

TY - THES

T1 - Computational inference and control of quality in multimedia services

AU - Menkovski, V.

PY - 2013

Y1 - 2013

N2 - Quality is the degree of excellence we expect of a service or a product. It is also one of the key factors that determine its value. For multimedia services, understanding the experienced quality means understanding how the delivered delity, precision and reliability correspond to the users' expectations. Yet the quality of multimedia services is inextricably linked to the underlying technology. It is developments in video recording, compression and transport as well as display technologies that enables high quality multimedia services to become ubiquitous. The constant evolution of these technologies delivers a steady increase in performance, but also a growing level of complexity. As new technologies stack on top of each other the interactions between them and their components become more intricate and obscure. In this environment optimizing the delivered quality of multimedia services becomes increasingly challenging. The factors that aect the experienced quality, or Quality of Experience (QoE), tend to have complex non-linear relationships. The subjectively perceived QoE is hard to measure directly and continuously evolves with the user's expectations. Faced with the diculty of designing an expert system for QoE management that relies on painstaking measurements and intricate heuristics, we turn to an approach based on learning or inference. The set of solutions presented in this work rely on computational intelligence techniques that do inference over the large set of signals coming from the system to deliver QoE models based on user feedback. We furthermore present solutions for inference of optimized control in systems with no guarantees for resource availability. This approach oers the opportunity to be more accurate in assessing the perceived quality, to incorporate more factors and to adapt as technology and user expectations evolve. In a similar fashion, the inferred control strategies can uncover more intricate patterns coming from the sensors and therefore implement farther-reaching decisions. Similarly to natural systems, this continuous adaptation and learning makes these systems more robust to perturbations in the environment, longer lasting accuracy and higher eciency in dealing with increased complexity. Overcoming this increasing complexity and diversity is crucial for addressing the challenges of future multimedia system. Through experiments and simulations this work demonstrates that adopting an approach of learning can improve the sub jective and objective QoE estimation, enable the implementation of ecient and scalable QoE management as well as ecient control mechanisms.

AB - Quality is the degree of excellence we expect of a service or a product. It is also one of the key factors that determine its value. For multimedia services, understanding the experienced quality means understanding how the delivered delity, precision and reliability correspond to the users' expectations. Yet the quality of multimedia services is inextricably linked to the underlying technology. It is developments in video recording, compression and transport as well as display technologies that enables high quality multimedia services to become ubiquitous. The constant evolution of these technologies delivers a steady increase in performance, but also a growing level of complexity. As new technologies stack on top of each other the interactions between them and their components become more intricate and obscure. In this environment optimizing the delivered quality of multimedia services becomes increasingly challenging. The factors that aect the experienced quality, or Quality of Experience (QoE), tend to have complex non-linear relationships. The subjectively perceived QoE is hard to measure directly and continuously evolves with the user's expectations. Faced with the diculty of designing an expert system for QoE management that relies on painstaking measurements and intricate heuristics, we turn to an approach based on learning or inference. The set of solutions presented in this work rely on computational intelligence techniques that do inference over the large set of signals coming from the system to deliver QoE models based on user feedback. We furthermore present solutions for inference of optimized control in systems with no guarantees for resource availability. This approach oers the opportunity to be more accurate in assessing the perceived quality, to incorporate more factors and to adapt as technology and user expectations evolve. In a similar fashion, the inferred control strategies can uncover more intricate patterns coming from the sensors and therefore implement farther-reaching decisions. Similarly to natural systems, this continuous adaptation and learning makes these systems more robust to perturbations in the environment, longer lasting accuracy and higher eciency in dealing with increased complexity. Overcoming this increasing complexity and diversity is crucial for addressing the challenges of future multimedia system. Through experiments and simulations this work demonstrates that adopting an approach of learning can improve the sub jective and objective QoE estimation, enable the implementation of ecient and scalable QoE management as well as ecient control mechanisms.

U2 - 10.6100/IR751521

DO - 10.6100/IR751521

M3 - Phd Thesis 1 (Research TU/e / Graduation TU/e)

SN - 978-90-386-3355-8

PB - Technische Universiteit Eindhoven

CY - Eindhoven

ER -