Network Representation Learning embeds each node in a network into a low-dimensional real-value vector which can be used for downstream tasks such as link prediction and recommendation. Many existing approaches use unsupervised or (semi-)supervised methods to explore the network topology and learn representations from it. In contrast, we propose, reinforcement learning network representations (RLNet), which explores the idea of using reinforcement learning to learn to explore the network and to obtain network representations. Based on reward signals, RLNet learns an actor which uses a policy to determine the network navigation actions. RLNet uses node representations to parameterize its policy, and the representations are learned together with the policy. Through experiments based on multiple datasets, we show that RLNet can obtain promising results in link prediction tasks.

Automated program repair recently received considerable attentions, and many techniques on this research area have been proposed. Among them, two genetic-programming-based techniques, GenProg and Par, have shown the promising results. In particular, GenProg has been used as the baseline technique to check the repair effectiveness of new techniques in much literature. Although GenProg and Par have shown their strong ability of fixing real-life bugs in nontrivial programs, to what extent GenProg and Par can benefit from genetic programming, used by them to guide the patch search process, is still unknown.

Researchers have proposed many spectra-based fault localization (SBFL) techniques in the past decades. Existing studies evaluate the effectiveness of these techniques from the viewpoint of developers, and have drawn some important conclusions through either empirical study or theoretical analysis. In this paper, we present the first study on the effectiveness of SBFL techniques from the viewpoint of fully automated debugging including the program repair of automation, for which the activity of automated fault localization is necessary. We assess the accuracy of fault localization according to the repair effectiveness in the automated repair process guided by the localization technique. Our experiment on 14 popular SBFL techniques with 11 subject programs shipping with real-life field failures presents the evidence that some conclusions drawn in prior studies do not hold in our experiment. Based on experimental results, we suggest that Jaccard should be used with high priority before some more effective SBFL techniques specially proposed for automated program repair occur in the future.

Many techniques on automated fault localization (AFL) have been introduced to assist developers in debugging. Prior studies evaluate the localization technique from the viewpoint of developers: measuring how many benefits that developers can obtain from the localization technique used when debugging. However, these evaluation approaches are not always suitable, because it is difficult to quantify precisely the benefits due to the complex debugging behaviors of developers. In addition, recent user studies have presented that developers working with AFL do not correct the defects more efficiently than ones working with only traditional debugging techniques such as breakpoints, even when the effectiveness of AFL is artificially improved. In this paper we attempt to propose a new research direction of developing AFL techniques from the viewpoint of fully automated debugging including the program repair of automation, for which the activity of AFL is necessary. We also introduce the NCP score as the evaluation measurement to assess and compare various techniques from this perspective. Our experiment on 15 popular AFL techniques with 11 subject programs shipping with real-life field failures presents the evidence that these AFL techniques performing well in prior studies do not have better localization effectiveness according to NCP score. We also observe that Jaccard has the better performance over other techniques in our experiment.

Recent techniques for fault localization statistically analyze coverage information of a set of test runs to measure the correlations between program entities and program failures. However, coverage information cannot identify those program entities whose execution affects the output, which weakens the aforementioned correlations. Thus, this paper proposes a novel statistical fault localization approach to address this problem. Our statistical approach utilizes program slices of a set of test runs to capture the influence of a program entity's execution on the output, and uses statistical analysis to measure the suspiciousness of program entities to be faulty. In addition, this paper presents a new slicing approach called approximate dynamic backward slice to balance the size and accuracy of a slice, and applies this slice to our statistical approach. The experimental results on two standard benchmarks show that our statistical approach significantly outperforms eight representative fault localization techniques.