Complicated class dependencies in object-oriented systems challenge traditional integration testing methods. Given varying efforts required to construct test stubs, determining optimal class test orders to minimize stubbing complexity is critical in integration testing. Reinforcement learning (RL), with its strengths in solving complex optimizations, has attracted substantial research attention. To fill the gap in the existing research that lacks of cross-strategy comparisons of RL, we systematically compare five RL algorithms’ performance using varied stubbing complexity weighting methods and critical class identification. Experimental results indicate that when fixed weights are used to calculate stubbing complexity, D3QN achieved the minimum overall stubbing complexity in seven out of nine tested programs. However, PPO attained the lowest mean overall stubbing complexity in five programs, demonstrating the best overall performance. DQN exhibited improved performance when stub complexity was calculated using the entropy-weighted method. After incorporating importance scores as a factor, D3QN achieved the best performance. The RL reward function with introduced importance values is more suitable for systems that are highly centralized and rely on core classes. This work provides robust empirical support and selection guidelines for the practical application of RL algorithms in CITO generation, as well as targeted insights for optimizing RL applications in this field.

Software systems are getting larger and more complex than ever before. In order to improve software reliability, software defect prediction is applied to assist developers in bug discovery. The ranking-oriented software defect prediction aims to rank software modules according to the predicted defect counts. However, existing ranking-oriented defect prediction models are constructed based on traditional hand-crafted features, which might overlook the rich syntactic information buried inside the source codes. In this paper, we propose a universal deep learning-based framework called United Deep Network for ranking-oriented software defect prediction. This framework utilizes deep neural networks to automatically generate features from source code with the syntactic and structural information preserved, and it can combine extracted features with traditional hand-crafted features in order to take advantage of both kinds of features to construct prediction models. Experimental results over 29 sets of data show the good performance of the proposed framework for building ranking-oriented defect prediction models.

Software defect prediction constructs a model from the previous version of a software project to predict defects in the current version, which can help software testers to focus on software modules with more defects in the current version. Most existing methods construct defect prediction models through minimizing the defect prediction error measures. Some researchers proposed model construction approaches that directly optimized the ranking performance in order to achieve an accurate order. In some situations, the model complexity is also considered. Therefore, defect prediction can be seen as a multi-objective optimization problem and should be solved by multi-objective approaches. And hence, in this paper, we employ an existing multi-objective evolutionary algorithm and propose a new multi-objective learning method based on it, to construct defect prediction models by simultaneously optimizing more than one goal. Experimental results over 30 sets of cross-version data show the effectiveness of the proposed multi-objective approaches.

Software defect prediction can help software testers to focus on software modules with more defects.Many ensemble methods have been proposed for software defect prediction to divide software modules into defect-prone and defect-free, and these ensemble methods have been proved to be more effective than single learning algorithms.A few ensemble approaches have been applied to predict the number of defects in software modules, and they also perform well in most cases.The good performance of ensemble approaches implies that ensemble algorithms might not only improve the accuracy of software defect classification models, but also improve the performance of defect ranking models.Therefore, we propose an ensemble method based on Yang et al.'s learning-to-rank approach in this paper.Experimental results show that the learning-to-rank-based ensemble approach performs better than the single learningto-rank approach, which means that the idea of ensemble can improve the performance of the learning-to-rank approach to sort modules in order of defect count.We also conduct a comparison study of ensemble approaches for cross-version defect prediction over 30 sets of cross-version data, which indicates that the ensemble technique of random subspace is more appropriate than boosting over these experimental data sets.

Empowered by today's rich media generating devices and convenient Internet access, crowdsourced live streaming (CSLS) service has developed rapidly and become one of the most popular Internet services. Large crowdsourced live streaming providers (CSLSPs) are migrating their services to geo-distributed cloud platforms (GDCPs) for lower costs and higher availability. A CSLSP may rent compute and network resources from cloud providers for video transcoding, video delivering, user-requests handling, and other related tasks. However, due to dynamic requests by viewers and widely spread locations of broadcasters and viewers, it is still challenging for a CSLSP to serve demands of users with reasonable resources from the cloud-based geo-distributed data centers. To overcome this challenge cost-effectively, we propose an online algorithm to save operational costs for CSLSPs by jointly and dynamically choosing right data centers for broadcasters and viewers. Mathematical analysis is presented and proves that our proposed online algorithm can ensure operational costs to be within an upper bound above the optimal solution, while guaranteeing the QoE for viewers. We conduct extensive trace-driven illustrative studies and show that the proposed method can achieve suboptimal results and outperforms other alternative methods.

Mobile edge computing (MEC) is a promising paradigm to integrate computing and communication resources in mobile networks. MEC can improve mobile service quality and enhance Quality of Experience (QoE) by offloading computation tasks to MEC servers. However, a MEC server only can provide limited computational resources for users. In this paper, we consider a mobile edge computing system that provides three offloading policies that are: (i) executing tasks in local device, (ii) offloading tasks to servers in a local region, (iii)offloading tasks to servers in a nearby region. In the policy (iii), mobile user equipment can utilize computational resources of MEC servers in nearby regions to solve the problem of insufficient computational resources in local region servers. We formulate the computation offloading problem as a potential game and propose a Distributed Offloading strategy based on Jacobi algorithm (DOJ) for solving the computation offloading problem in a short period. The simulation results show that our proposed algorithm can reduce overall system costs and guarantee the QoE of users.

Dynamic requests of viewers from sparse and dispersed locations for crowdsourced-live-streaming (CSLS) service make current cloud service providers (CSPs) inadequate to provide sufficient quality of experience (QoE). To solve this issue, we propose a multi-CDN-assisted-CSLS (MCACLS) architecture, a novel cloud architecture complemented by multiple content delivery networks (Multi-CDNs). MCACLS architecture can enhance a CSP's capacity of video distribution service and improve the quality of CSLS service for end-users while reducing the overall operational cost. MCACLS adaptively adjusts resources between a CSP and its leased CDN service in a fine granularity to deal with the volatility of user requests. However, scheduling resources cost-effectively in response to user requests from different regions is a critical issue that must be addressed. We formulate the above problem into a constrained stochastic optimization problem and propose an algorithm based on the Nash bargaining solution. Our proposed algorithm makes tradeoff between QoE of users and the overall operational cost for CSPs. Illustrative studies validate the advantages of MCACLS and show that it is more cost-effective, reducing the overall operational cost by up to 15% compared with other alternatives while achieving sufficient QoE for viewers.

Sorting software modules in order of defect count can help testers to focus on software modules with more defects. One of the most popular methods for sorting modules is generalized linear regression. However, our previous study showed the poor performance of these regression models, which might be caused by severe multicollinearity. Ridge regression (RR) can improve the prediction performance for multicollinearity problems. Lasso regression (LAR) is a worthy competitor to RR. Therefore, we investigate both RR and LAR models for cross-version defect prediction. Cross-version defect prediction is an approximate to real applications. It constructs prediction models from a previous version of projects and predicts defects in the next version. Experimental results based on 11 projects from the PROMISE repository consisting of 41 different versions show that: 1) there exist severe multicollinearity problems in the experimental datasets; 2) both RR and LAR models perform better than linear regression and negative binomial regression for cross-version defect prediction; and 3) compared with two best methods in our previous study for sorting software modules according to the predicted number of defects, RR has comparable performance and less model construction time.

Software defect prediction can help to allocate testing resources efficiently through ranking software modules according to their defects. Existing software defect prediction models that are optimized to predict explicitly the number of defects in a software module might fail to give an accurate order because it is very difficult to predict the exact number of defects in a software module due to noisy data. This paper introduces a learning-to-rank approach to construct software defect prediction models by directly optimizing the ranking performance. In this paper, we build on our previous work, and further study whether the idea of directly optimizing the model performance measure can benefit software defect prediction model construction. The work includes two aspects: one is a novel application of the learning-to-rank approach to real-world data sets for software defect prediction, and the other is a comprehensive evaluation and comparison of the learning-to-rank method against other algorithms that have been used for predicting the order of software modules according to the predicted number of defects. Our empirical studies demonstrate the effectiveness of directly optimizing the model performance measure for the learning-to-rank approach to construct defect prediction models for the ranking task.