Our department celebrates publication of a new paper entitled “"Maximum Entropy Discrimination Poisson Regression for Software Reliability Modeling” in the IEEE Transactions on Neural Networks and Learning Systems. The IEEE Transactions on Neural Networks and Learning Systems is the most significant journal in the field of Neural Networks and Learning Systems research, and the second most prestigious journal in the general field of Machine Learning.

Authors of this work are two faculty members of our department, namely Dr. Sotirios P. Chatzis and Dr. Andreas Andreou. In this work, the authors introduce a hierarchical Bayesian model for regression modeling of counts, that leverages the strengths of the max-margin principle to learn the function mapping software metrics to predicted counts of bugs. Specifically, the proposed model is based on a doubly stochastic homogeneous Poisson process construction, where the failure rate parameter at each time point is modeled through a mixture prior of max-margin regression models. Application of the max-margin learning principle allows for obtaining a more discriminative learning technique, making more effective use of our training data during inference. In addition, the utilization of a mixture of max-margin regression priors, each one modeling a different (latent) subspace of the parameter space, allows for learning multimodal underlying data distributions with increased flexibility compared to single-component models. As shown in the paper, the advanced statistical machine learning machinery devised in the context of the proposed model allows for yielding state-of-the-art defect prediction performance in several challenging benchmark software systems.

This paper is a characteristic example of successful interdisciplinary research taking place in our department, built upon our core competencies (unique not only within Cyprus but also in SE Europe), which revolve around Data Science and Engineering.