A Machine Learning Resource Allocation Solution to Improve Video Quality in Remote Education
Abstract : The current global pandemic crisis has unquestionably disrupted the higher education sector, forcing educational institutions to rapidly embrace technology-enhanced learning. However, the COVID-19 containment measures that forced people to work or stay at home, have determined a significant increase in the Internet traffic that puts tremendous pressure on the underlying network infrastructure. This affects negatively content delivery and consequently user perceived quality, especially for video-based services. Focusing on this problem, this paper proposes a machine learning-based resource allocation solution that improves the quality of video services for increased number of viewers. The solution is deployed and tested in an educational context, demonstrating its benefit in terms of major quality of service parameters for various video content, in comparison with existing state of the art. Moreover, a discussion on how the technology is helping to mitigate the effects of massively increasing internet traffic on the video quality in an educational context is also presented.
For existing wireless systems assisted by cloud computing, a huge amount of data on historical scenarios may have been collected and stored at the cloud. The strong computing capability of the cloud is exploited to search the optimal or near-optimal solutions for these historical scenarios. By classifying these solutions, the similarities hidden in these historical scenarios are extracted as a machine learning based resource allocation scheme. The machine learning based resource allocation scheme will be forwarded to guide BS how to allocate radio resource more efficiently. When a BS is deployed in a new area, there is usually no available data about historical scenarios.
? The proposed controller from Fig. 5 is employed to find good solutions at each TTI in terms of the prioritization scheme and scheduling rules to be used for each video class.
? Based on the scheduler-controller interaction, these decisions can be learnt over time to maximize as much as possible the QoS provisioning for all traffic classes.
? However, due to the high dimension of the scheduler state space, these pairs cannot be exhaustively enumerated and hence, the optimal decisions can be only approximated. Thus, the aim is to learn the parameterization of some non-linear functions to approximate the best decisions afferent to sub-problems a
The proposed machine learning framework is shown in Fig. 2. At the cloud, a huge amount of historical data on scenarios are stored using the cloud storage. The historical data has a lot of attributes, including the user number, the CSI of users, international mobile subscriber identification numbers (IMSIs) of users, and so on. Some attributes, such as IMSIs of users, may be irrelevant for the specific resource allocation, i.e., these irrelevant attributes are not included in the parameter ve 7 problem (1). Learning from a large number of raw data with many attributes generally requires a large amount of memory and computation power, and it may influence the learning accuracy .
? Adaptive rich media delivery solutions help increase user QoE in general, and in the specific context of learning, research studies show that they also have potential to increase learner QoE and academic performance.
? Compared to other TDP strategies where users from different classes may be pre-selected according to their QoS budget [32], [33], the proposed scheduler aims to prioritize classes by deciding at each TTI a new prioritization sequence.
? However, the order of classes to be scheduled at each TTI is decided based on the performance of x over the requirement x¯ and not on the occupancy degree of the available spectrum.
? As a consequence, depending on the traffic load, some classes may remain unscheduled since all RBs are allocated to learners from classes with higher priorities decided at TTI t.
|