Active Content Popularity Learning and Caching Optimization With Hit Ratio Guarantees

Abstract

Edge caching is an effective solution to reduce delivery latency and network congestion by bringing contents close to end-users. A deep understanding of content popularity and the principles underlying the content request sequence are required to effectively utilize the cache. Most existing works design caching policies based on global content requests with very limited consideration of individual content requests which reflect personal preferences. To enable the optimal caching strategy, in this paper, we propose an Active learning (AL) approach to learn the content popularities and design an accurate content request prediction model. We model the content requests from user terminals as a demand matrix and then employ AL-based query-by-committee (QBC) matrix completion to predict future missing requests. The main principle of QBC is to query the most informative missing entries of the demand matrix. Based on the prediction provided by the QBC, we propose an adaptive optimization caching framework to learn popularities as fast as possible while guaranteeing an operational cache hit ratio requirement. The proposed framework is model-free, thus does not require any statistical knowledge about the underlying traffic demands. We consider both the fixed and time-varying nature of content popularities. The effectiveness of the proposed learning caching policies over the existing methods is demonstrated in terms of root mean square error, cache hit ratio, and cache size on a simulated dataset. INDEX TERMS Edge caching, Active learning, Matrix completion, Content popularity.

A Comprehensive Review on Edge Caching from the Perspective of Total Process: Placement, Policy and Delivery

“…Paper [ 94 ] is based on the research of paper [ 93 ], adding collaborative caching between BSs to further improve user experience. Changes in content popularity are often unstable, and the selected feature samples also need to be continuously updated [ 95 ]. Tang et al [ 96 ] proposed a reinforcement learning-based scheme to reflect the popularity of files and user preferences, in which training samples are continuously generated through the feedback mechanism of the Markov Decision Process (MDP).…”

Learning to Cache: Federated Caching in a Cellular Network With Correlated Demands

“…Moreover, in order to use spatio-temporal information, the authors in [32] developed a Bayesian dynamical model to predict the popularity and minimized the cost for transferring data among the sBSs in the network. On the other hand, in [33], active learning approach is used to learn the content popularities to design an accurate content request prediction model. In [34], the authors propose joint caching and the dynamic multicast scheduling to increase the robustness of wireless transmission.…”

“…t R b,T +1 (π b,t ) ≥ T t=T −τ α b,t R b,T +1 (C * b,t ) Reg b,T,τ (π b,t ) τ . (36)From(33), we haveE[R b,T +1 (π (av) b,T +1 )|Z T b,1 ] ≥ T t=T −τ α b,t R b,T +1 (C * b,t ) Reg b,T,τ (π b,t ) τ −H max ∥α b,T ∥ 2 2 τ log 1 δ − M b,T +1 (w ̸ =b,T ) − D b,T (α b,T,τ ). +1 , we get E[R b,T +1 (π (av) b,T +1 )|Z T b,1 ] ≥ E R b,T +1 (C * b,T +1 )|Z T 2Reg b,T,τ (π b,t ) τ −2H max ∥α b,T ∥ 2 2 τ log 1 δ − M b,T +1 (w ̸ =b,T ) − 2D b,T (α b,T,τ ).…”

“…(38)It is possible to choose C * b,t in such as way thatE R b,T +1 (C * b,T +1 ) t E R b,T +1 (h b,t ) | Z T b,1 − γfor some γ > 0. Using this in the above equation, and substituting the resulting equation in(33) gives E R b,T +1 (π(av) b,T +1 ) | Z T b,1…”

Trend-Aware Proactive Caching via Tensor Train Decomposition: A Bayesian Viewpoint

“…For example, content popularity can spread through word-of-mouth across social connections. Most of prior work on content request prediction focused mainly on static scenario and ignored the time evolution pattern in the requests, e.g., [6]- [8]. Content requests, however, exhibit time-dependency which can have considerable effect on the caching performance.…”