EURASIP Journal on Wireless Communications and Networking

Clustering, a traditional machine learning method, plays a significant role in data analysis. Most clustering algorithms depend on a predetermined exact number of clusters, whereas, in practice, clusters are usually unpredictable. Although the Elbow method is one of the most commonly used methods to discriminate the optimal cluster number, the discriminant of the number of clusters depends on the manual identification of the elbow points on the visualization curve. Thus, experienced analysts cannot clearly identify the elbow point from the plotted curve when the plotted curve is fairly smooth. To solve this problem, a new elbow point discriminant method is proposed to yield a statistical metric that estimates an optimal cluster number when clustering on a dataset. First, the average degree of distortion obtained by the Elbow method is normalized to the range of 0 to 10. Second, the normalized results are used to calculate the cosine of intersection angles between elbow points. Third, this calculated cosine of intersection angles and the arccosine theorem are used to compute the intersection angles between elbow points. Finally, the index of the above-computed minimal intersection angles between elbow points is used as the estimated potential optimal cluster number. The experimental results based on simulated datasets and a well-known public dataset (Iris Dataset) demonstrated that the estimated optimal cluster number obtained by our newly proposed method is better than the widely used Silhouette method.

Mobile edge computing (MEC) emerges recently as a promising solution to relieve resource-limited mobile devices from computation-intensive tasks, which enables devices to offload workloads to nearby MEC servers and improve the quality of computation experience. In this paper, an MEC enabled multi-user multi-input multi-output (MIMO) system with stochastic wireless channels and task arrivals is considered. In order to minimize long-term average computation cost in terms of power consumption and buffering delay at each user, a deep reinforcement learning (DRL)-based dynamic computation offloading strategy is investigated to build a scalable system with limited feedback. Specifically, a continuous action space-based DRL approach named deep deterministic policy gradient (DDPG) is adopted to learn decentralized computation offloading policies at all users respectively, where local execution and task offloading powers will be adaptively allocated according to each user’s local observation. Numerical results demonstrate that the proposed DDPG-based strategy can help each user learn an efficient dynamic offloading policy and also verify the superiority of its continuous power allocation capability to policies learned by conventional discrete action space-based reinforcement learning approaches like deep Q-network (DQN) as well as some other greedy strategies with reduced computation cost. Besides, power-delay tradeoff for computation offloading is also analyzed for both the DDPG-based and DQN-based strategies.