Deep State Inference: Toward Behavioral Model Inference of Black-Box Software Systems

Many software engineering tasks, such as testing, debugging, and anomaly detection can benefit from the ability to infer a behavioral model of software. Most existing inference approaches assume access code collect execution sequences. In this paper, we investigate black-box scenario, where system under analysis cannot be instrumented in fashion. This scenario is particularly common when it comes control logs, which often take form continuous signals. situation, an trace amounts multivariate time-series input output signals, different states correspond “phases” time-series. From perspective, challenge detect these phase changes place. Unfortunately, most solutions are either univariate, make assumptions about data distribution, or have limited learning power. paper propose hybrid deep neural network that accepts time series applies set convolutional recurrent layers learn non-linear correlations between signals patterns over time. We show how approach used accurately state changes, inferred models successfully applied transfer-learning scenarios, process traces products with similar characteristics. Our experimental results on two UAV autopilot case studies (one industrial one open-source) indicate our highly accurate (over 90% F1 score for classification) significantly improves baselines (by up 102% change point detection). Using transfer also maximum achievable scores open-source study achieved by reusing trained only fine tuning them using low 5 labeled samples, reduces manual labeling effort 98%.