Providing natural language-based explanations to justify recommendations helps to improve users' satisfaction and gain users' trust. However, as current explanation generation methods are commonly trained with an objective to mimic existing user reviews, the generated explanations are often not aligned with the predicted ratings or some important features of the recommended items, and thus, are suboptimal in helping users make informed decision on the recommendation platform. To tackle this problem, we propose a flexible model-agnostic method named MMI (Maximizing Mutual Information) framework to enhance the alignment between the generated natural language explanations and the predicted rating/important item features. Specifically, we propose to use mutual information (MI) as a measure for the alignment and train a neural MI estimator. Then, we treat a well-trained explanation generation model as the backbone model and further fine-tune it through reinforcement learning with guidance from the MI estimator, which rewards a generated explanation that is more aligned with the predicted rating or a pre-defined feature of the recommended item. Experiments on three datasets demonstrate that our MMI framework can boost different backbone models, enabling them to outperform existing baselines in terms of alignment with predicted ratings and item features. Additionally, user studies verify that MI-enhanced explanations indeed facilitate users' decisions and are favorable compared with other baselines due to their better alignment properties.

This paper proposes Personalized Diversity-promoting GAN (PD-GAN), a novel recommendation model to generate diverse, yet relevant recommendations. Specifically, for each user, a generator recommends a set of diverse and relevant items by sequentially sampling from a personalized Determinantal Point Process (DPP) kernel matrix. This kernel matrix is constructed by two learnable components: the general co-occurrence of diverse items and the user's personal preference to items. To learn the first component, we propose a novel pairwise learning paradigm using training pairs, and each training pair consists of a set of diverse items and a set of similar items randomly sampled from the observed data of all users. The second component is learnt through adversarial training against a discriminator which strives to distinguish between recommended items and the ground-truth sets randomly sampled from the observed data of the target user. Experimental results show that PD-GAN is superior to generate recommendations that are both diverse and relevant.

The performance and reliability of services relies on the network virtualization environment's capabilities to effectively detect and diagnose faults in both substrate and virtual network. However, Network Virtualization Environment (NVE) brings to fault diagnosis new challenges such as inaccessible substrate network information and multi-layer faults. To solve the above issues, a Multi-layer Fault Diagnosis Method (MFDM) is proposed. A layer-by-layer strategy is used to resolve the problem of inaccessible substrate network information. And a filtering algorithm is proposed to distinguish the multi-layer faults in the network virtualization environment. At last, a contribution-based hypothesis selection algorithm is proposed to infer the most possible faults. Simulations and experimental results show that MFDM has a higher performance in the accuracy ratio, false-positive ratio.

Adaptive active diagnosis method is widely adopted for fault diagnosis in networks. In active diagnosis, appropriate probes are selected sequentially and fault diagnosis is made by inference from results of selected probes. It is very important to select active probes with low cost and less impact on network performance. However, the selection of the most informative set of probes with limited cost is an NP-hard problem. The computational complexities of existing probe selection algorithms are still too high for large scale networks. In this paper, a lemma about mutual information provided by probes is proved based on the property of conditional entropy. Then an approximate method derived from this lemma is introduced to compute mutual information of probe. With this approximate method an efficient probe selection algorithm for active diagnosis is proposed. At last, the efficiency and effectiveness of the proposed algorithm is verified through simulation.

Reducing the number of active nodes is an natural and effective method in energy saving for large scale cloud datacenters. However, the traditional way of consolidating VMs by migration may lead to the impacts on performance as well as costs on power consumption and migration. In this paper, we propose an algorithm named DRAMDT to deploy VMs to servers without the use of migration, which groups VM instances by their deadline runtimes and pack VMs within the same group on the same servers. It is demonstrated in the experiments that DRAMDT is able to shut down more servers in time to save energy when user amount reduces. And the proposed algorithm has small searching space, low complexity and good scalability.

Energy efficiency of cloud computing attracts a great deal of attention recently more than ever. In this paper, based on the characteristics of virtualized cloud environment and MapReduce workload, an energy efficiency management method called EECLOUD is presented. The method involves two strategies: VM allocation and dynamic VM placement strategies. The VM allocation strategy aims at maximizing an energy utility function making trade-off between job runtime and allocated virtual resources. And the VM placement strategy makes energy-aware scheme to place VMs on servers taking into account both VM duration and resource capacity. Additionally, a dynamic VM-host mapping algorithm is proposed to place VMs to the least amount of physical server, which uses live migration to enhance capacity utilization. The experimental results show that our approach achieves greater energy savings over existing methods.

Energy consumption has become major operational cost in data centers. Virtualization technology used in cloud computing platforms can improve energy efficiency and reduce costs. There are many ongoing research projects focusing on power management for virtualized cloud by making power-aware resource allocation and scheduling policies. However, there is a lack of VM power profiling method in such research, because the power consumption of an individual virtual machine (VM) cannot be measured directly by hardware power meter. In this paper, a novel power metering model is proposed for VMs in the cloud environment, based on online monitoring of system resource metrics, to estimate the power consumption of a physical server as well as one or more VMs running on it. By analyzing problems found in experiments, the model is improved to be the classified-piecewise ternary linear regression model which can achieve higher accuracy. In addition, the model is proved to be effective by running a variety of sample programs. The implementation of our model shows that it can achieve average estimation accuracy of more than 96% with low runtime overhead.

In [1], a modified Vertical-Bell Labs layered space- time (V-BLAST) system has been proposed, where different delay offsets are intentionally applied to the spatially multiplexed data streams. Compared with the conventional synchronous V- BLAST system, the modified system with only a simple linear detection under zero-forcing(ZF) criterion can achieve Mrth- order diversity(where Mris the number of receive antennas), therefore giving a reasonable tradeoff between complexity and performance. In this paper, we focus on how to find the optimum linear detection of the Modified V-BLAST Systems. Simulations demonstrate that the optimum linear detection is superior to the ZF detection algorithm used in [Shihai Shao et al., 2007], especially when the number of transmit antennas is large.