#### DMCA

## Learning-based optimization of cache content in a small cell base station,” arXiv preprint arXiv:1402.3247 (2014)

Citations: | 5 - 0 self |

### Citations

1063 | Web caching and Zipf-like distributions: evidence and implications
- BRESLAU, CAO, et al.
- 1999
(Show Context)
Citation Context ...ent users. Furthermore, we assume that the popularity profile of the files is characterized by a Zipflike distribution with parameter γ, which is commonly used to model content popularity in networks =-=[9]-=-. Hence, the popularity of file f , that is, the expected number of requests in period t, is θf = Ufγ∑Fi=1 1iγ , and the content popularity profile is Θ = (θ1, . . . , θF ). Notice that parameter γ mo... |

785 | Finite-time analysis of the multiarmed bandit problem
- Auer, Cesa-Bianchi, et al.
- 2002
(Show Context)
Citation Context ...o find a policy with a small regret. In [10], Lai and Robbins show that no policy can achieve an asymptotic regret smaller than O(log(t)), where O(·) includes a multiplicative and a constant term. In =-=[11]-=- a policy, called the upper confidence bounds (UCB), is presented, and proven to achieve a regret on the order of O(log(t)), uniformly over t. B. MAB for optimal caching: Regret bound We will first il... |

497 |
Asymptotically efficient adaptive allocation rules
- Lai, Robbins
- 1985
(Show Context)
Citation Context ...t of the policy that always pulls the best arm. Hence, the regret is a measure of the loss due to not knowing the reward profile of the arms. The objective is to find a policy with a small regret. In =-=[10]-=-, Lai and Robbins show that no policy can achieve an asymptotic regret smaller than O(log(t)), where O(·) includes a multiplicative and a constant term. In [11] a policy, called the upper confidence b... |

109 |
Regret analysis of stochastic and nonstochastic multi-armed bandit problems
- Bubeck, Bianchi
- 2012
(Show Context)
Citation Context ... traffic served from the sBS without knowing the popularity profile in advance, and by observing only the requests corresponding to the files in the cache. We model this as a multi-armed bandit (MAB) =-=[8]-=- problem, and provide several caching algorithms. To the best of the authors’ knowledge this is the first paper to address the optimal caching problem in the case of unknown popularity profiles using ... |

73 | Distributed caching algorithms for content distribution networks
- Borst, Gupta, et al.
- 2010
(Show Context)
Citation Context ...ant problem is to decide which content should be cached in the limited storage space available at the sBS. Cache content placement has been studied for different cache topologies and applications. In =-=[6]-=-, a cache cluster formed by several leaf-cache nodes and a parent-cache node is studied. Files requested by users connected to a leafcache node are fetched either directly from the leaf-cache node, fr... |

66 |
Discrete-variable extremum problems
- Dantzig
- 1957
(Show Context)
Citation Context .../wF , it can be estimated that δ ≤ 2 [16]. Hence, the greedy algorithm is an (α, β)-solver with α = 0.5 and β = 1. For the general knapsack problem, the regularity condition ensures that Lemma 1 hold =-=[17]-=-. If Θ characterises the Zipf distribution, we can obtain a tighter bound for δ: δ ≤ 1 + s1 s|S| 1/ ⌈ M s1 ⌉γ ∑⌊Ms1 ⌋ i=1 1/iγ . (5) If the cache size is much larger than the maximum file size, i.e., ... |

56 | FemtoCaching: Wireless Video Content Delivery through Distributed Caching Helpers,” submitted for publication, available on http://arxiv.org/pdf/1109.4179v1
- Golrezaei, Shanmugam, et al.
(Show Context)
Citation Context ...maximising the QoE of the users that are connected to it by allocating some of the available bandwidth to serve its users, and trading the rest with other caches in an auction-game fashion. Reference =-=[4]-=- studies a caching problem in a backhaul-constrained small cell network, in which users can connect to several sBSs, and fetch files from the sBS that offers the least latency. The problem is shown to... |

35 | Improved algorithms for linear stochastic bandits
- Abbasi-Yadkori, Pal, et al.
- 2011
(Show Context)
Citation Context ... pi until period t. A MAB problem in which several arms can be pulled simultaneously is known as a combinatorial MAB (CMAB) [12] if the individual reward for each arm is observed, and as a linear MAB =-=[13]-=- if only an aggregated reward is observed. In this paper we consider the CMAB problem, since we assume that the individual instantaneous demands of all files in the cache are observed. An algorithm th... |

34 | Integer-programming software systems
- Atamtürk, Savelsbergh
- 2005
(Show Context)
Citation Context ...s a knapsack problem with values vf = θfSf , and weights wf = Sf . Knapsack problems, which are known to be NP-hard, can be solved optimally using branching algorithms, such as branch and bound (B&B) =-=[15]-=-. B&B worst case complexity is exponential, same as exhaustive search. Approximate solutions are obtained with low-complexity algorithms. A. Approximate solution We consider a linear-relaxation intege... |

22 | Combinatorial multi-armed bandit: General framework, results and applications
- Chen, Wang, et al.
(Show Context)
Citation Context ...t) − racpi (t), where racpi (t) is the expected accumulated reward of policy pi until period t. A MAB problem in which several arms can be pulled simultaneously is known as a combinatorial MAB (CMAB) =-=[12]-=- if the individual reward for each arm is observed, and as a linear MAB [13] if only an aggregated reward is observed. In this paper we consider the CMAB problem, since we assume that the individual i... |

8 | Algorithms for multi-armed bandit problems
- Kuleshov, Precup
(Show Context)
Citation Context ...for example, obtained from the content provider. C. MAB for optimal caching: Application Despite the logarithmic growth of regret, CUCB can take many iterations to learn the optimal cache content. In =-=[14]-=- it is shown for the classic MAB problem that, much simpler algorithms than UCB can achieve higher performance in practice. One such simple algorithm is the -greedy algorithm, which caches at each it... |

7 | Collaborative caching in wireless video streaming through resource auctions
- Dai, Liu, et al.
- 2012
(Show Context)
Citation Context ...e (QoE) of the users and alleviate congestion in the sBS backhaul connection has received great attention from both the industry, with the development of cache-enabled sBSs [1], [2], and the academia =-=[3]-=-–[5]. Enabled by the drastic reduction in price and size of data-storage devices, significant amount of popular content, such as news feeds or YouTube videos, can be stored in sBSs, and reliably deliv... |

5 | Exploiting user mobility for wireless content delivery
- Poularakis, Tassiulas
- 2013
(Show Context)
Citation Context ...oE) of the users and alleviate congestion in the sBS backhaul connection has received great attention from both the industry, with the development of cache-enabled sBSs [1], [2], and the academia [3]–=-=[5]-=-. Enabled by the drastic reduction in price and size of data-storage devices, significant amount of popular content, such as news feeds or YouTube videos, can be stored in sBSs, and reliably delivered... |

2 |
www.saguna.net (Accessed: 14
- page
- 2013
(Show Context)
Citation Context ...se the quality of experience (QoE) of the users and alleviate congestion in the sBS backhaul connection has received great attention from both the industry, with the development of cache-enabled sBSs =-=[1]-=-, [2], and the academia [3]–[5]. Enabled by the drastic reduction in price and size of data-storage devices, significant amount of popular content, such as news feeds or YouTube videos, can be stored ... |

1 |
Optimal cooperative content placement algorithms in hierarchical cache topologies
- Poularakis, Tassiulas
(Show Context)
Citation Context ...acing content in the cache nodes in order to minimise the total bandwidth consumption is studied, and approximate solutions are given for special cases. The cache cluster in [6] is further studied in =-=[7]-=- for the case of equal size files and when leaf-cache nodes are not interconnected, and a polynomial complexity optimal algorithm is given for the case of two leaf-cache nodes. A caching problem with ... |

1 |
Exact and greedy solutions of the knapsack prblem: the ratio of values of the objective function
- Korbut, Sigal
- 2010
(Show Context)
Citation Context ...n the value of the optimal solution and that of the greedy approximation. If θi ≥ θj ,∀j < i, and (4) fulfils the regularity condition, that is, v1/w1 ≥ . . . ≥ vF/wF , it can be estimated that δ ≤ 2 =-=[16]-=-. Hence, the greedy algorithm is an (α, β)-solver with α = 0.5 and β = 1. For the general knapsack problem, the regularity condition ensures that Lemma 1 hold [17]. If Θ characterises the Zipf distrib... |