Citation Context

... . , Ftar}, where each content f has a size of L(f) and a bitrate requirement of B(f). Moreover, we suppose that users’ content requests follow a Zipf-like distribution PFtar (f),∀f ∈ Ftar defined as =-=[15]-=-: PFtar (f) = Ω fα (1) where Ω = (∑Ftar i=1 1 iα )−1 and α characterizes the steepness of the distribution, reflecting different content popularities. Having such a content popularity in the ordered c...

Citation Context

...3]. Given the fact that data sparsity and cold-start problems degrade the performance of proactive caching, we leverage the framework of transfer learning (TL) and recent advances in machine learning =-=[14]-=-. TL is motivated by the fact that in many real-world applications, it is hard or even impossible to collect and label training data to build suitable prediction models. Exploiting available data from...

Citation Context

...rred as the source domain in the sequel. Specifically, this source domain contains the behaviour of users’ interactions within their social communities, modelled as a Chinese restaurant process (CRP) =-=[16]-=-. This constitutes the prior information used in the transfer learning procedure. In the CRP with parameter β, every customer selects an occupied table with a probability proportional to the number of...

Citation Context

...ng, transfer learning, collaborative filtering, data sparsity, cold-start problem, 5G I. INTRODUCTION Caching at the network edge is one of the five most promising innovations in 5G wireless networks =-=[1]-=-. Recently, it was shown that caching can significantly offload different segments of the infrastructure including radio access network (RAN) and core network (CN), by intelligently storing contents c...

Citation Context

...ny-to-many matching [8] and joint contentaware user clustering and content caching [11]. Other works include information-theoretic studies looking at fundamentals of local and global caching gains in =-=[9]-=-, facility location based approximation in [7], as well as multiple-input multiple-output (MIMO) caching in [12], and coded caching in [10]. In [2], by exploiting spatio-social caching coupled with D2...

Citation Context

...rtheless, the content popularity matrix remains typically large and sparse with very few users ratings, rendering CF learning methods inefficient mainly due to data sparseness and cold-start problems =-=[13]-=-. Given the fact that data sparsity and cold-start problems degrade the performance of proactive caching, we leverage the framework of transfer learning (TL) and recent advances in machine learning [1...

Citation Context

...tents closer to the users. As opposed to pushing contents on a besteffort basis ignoring end-users’ behavior and interactions, we are witnessing an era of truly context-aware and proactive networking =-=[2]-=-. Undoubtedly, edge caching has taken recent 5G research activities by storm as evidenced by the recent literature in both academia and industry [2]–[12] (to cite a few). Although caching has been wel...

Citation Context

...pported by the ERC Starting Grant 305123 MORE (Advanced Mathematical Tools for Complex Network Engineering), the SHARING project under the Finland grant 128010 and the project BESTCOM. are studied in =-=[4]-=-, characterizing the outage probability and average delivery rate. A stochastic-geometry based caching framework for device-to-device (D2D) communications is examined in [5] where mathematical express...

Citation Context

...ns of served content requests are given. From a game theoretic standpoint, various approaches have been studied such as multi-armed bandits under unknown content popularity [6], many-to-many matching =-=[8]-=- and joint contentaware user clustering and content caching [11]. Other works include information-theoretic studies looking at fundamentals of local and global caching gains in [9], facility location ...

Citation Context

...of local and global fractions of served content requests are given. From a game theoretic standpoint, various approaches have been studied such as multi-armed bandits under unknown content popularity =-=[6]-=-, many-to-many matching [8] and joint contentaware user clustering and content caching [11]. Other works include information-theoretic studies looking at fundamentals of local and global caching gains...

Citation Context

... looking at fundamentals of local and global caching gains in [9], facility location based approximation in [7], as well as multiple-input multiple-output (MIMO) caching in [12], and coded caching in =-=[10]-=-. In [2], by exploiting spatio-social caching coupled with D2D communication, we proposed a novel proactive networking paradigm in which SBSs and user terminals (UTs) proactively cache contents at the...

Citation Context

...of truly context-aware and proactive networking [2]. Undoubtedly, edge caching has taken recent 5G research activities by storm as evidenced by the recent literature in both academia and industry [2]–=-=[12]-=- (to cite a few). Although caching has been well-studied in wired networks, caching over wireless remains in its infancy. The idea of femtocaching is proposed in [3], in which small base stations (SBS...

Citation Context

...ct BESTCOM. are studied in [4], characterizing the outage probability and average delivery rate. A stochastic-geometry based caching framework for device-to-device (D2D) communications is examined in =-=[5]-=- where mathematical expressions of local and global fractions of served content requests are given. From a game theoretic standpoint, various approaches have been studied such as multi-armed bandits u...

Citation Context

... user clustering and content caching [11]. Other works include information-theoretic studies looking at fundamentals of local and global caching gains in [9], facility location based approximation in =-=[7]-=-, as well as multiple-input multiple-output (MIMO) caching in [12], and coded caching in [10]. In [2], by exploiting spatio-social caching coupled with D2D communication, we proposed a novel proactive...

Citation Context

...standpoint, various approaches have been studied such as multi-armed bandits under unknown content popularity [6], many-to-many matching [8] and joint contentaware user clustering and content caching =-=[11]-=-. Other works include information-theoretic studies looking at fundamentals of local and global caching gains in [9], facility location based approximation in [7], as well as multiple-input multiple-o...