Novel developments in computational intelligence with applications to software quality

该提案将支持对计算智能和机器学习理论的新研究，以及它们在建模软件可靠性方面的应用。我们的主要理论贡献将在“复杂模糊逻辑”的新领域，这是最近提出的模糊逻辑的推广。复模糊逻辑是一种新颖的多值逻辑，其真值为复数。目前，人们对这种逻辑的性质以及如何应用所知甚少。我们的研究计划将通过发展复杂模糊逻辑的数学逻辑基础和将复杂模糊逻辑应用于实际问题的归纳学习算法来解决这两个问题。我们相信复杂模糊逻辑在时间序列预测中会特别有效。我们对机器学习的研究将集中在分层上，这是一种在机器学习数据集中减少偏度的方法。分层算法从多数类中删除样本和/或向少数类中添加样本，从而消除或减少偏度。我们将研究各种分层算法的概括，以及它们在商业智能中常用的高性能计算环境中的实现。我们的理论贡献将应用于软件可靠性的某些方面的建模。我们将研究分层算法在软件缺陷预测中的使用。实际上，在软件缺陷预测中使用分层的工作是不存在的。我们将通过广泛的实验调查，在这方面提供新的证据。软件可靠性的另一个方面是软件可靠性增长建模（SRGM）。在以前的工作中，我们已经研究了混沌理论在SRGM中的应用。我们将继续这项研究，我们也将把我们新的归纳学习算法应用于复杂模糊逻辑。这些方法是非常新颖的，我们期望我们的方法将显著改进当前的SRGM技术。我们的贡献将通过提供新的理论指导和一批在智能系统中训练的HQP，使加拿大智能系统行业受益。

This proposal will support new investigations into the theory of computational intelligence and machine learning, and their application to modeling software reliability. Our principal theoretical contributions will be in the new area of "complex fuzzy logic," a recently proposed generalization of fuzzy logic. Complex fuzzy logic is a novel multi-valued logic, whose truth values are complex numbers. Little is currently known about the properties of this logic, or how it might be applied. Our program of research will address both of these questions, by developing both the mathematical-logic foundations of complex fuzzy logic, and inductive learning algorithms that apply complex fuzzy logic to practical problems. We believe that complex fuzzy logic will be particularly effective for time series forecasting. Our research into machine learning will focus on stratification, a method for reducing skewness in machine learning datasets. Stratification algorithms delete samples from the majority class and/or add examples to the minority class, removing or reducing skewness. We will investigate generalizations of various stratification algorithms, and their implementation in the high-performance computing environments commonly used in business intelligence.  Our theoretical contributions will be applied to modeling certain aspects of software reliability. We will investigate the use of stratification algorithms for software defect prediction. There is virtually no existing work on the use of stratification in software defect prediction. We will provide new evidence in this area through extensive experimental investigation. Another aspect of software reliability is software reliability growth modeling (SRGM). In previous work we have investigated the use of chaos theory for SRGM. We will continue that investigation, and we will also apply our new inductive learning algorithms for complex fuzzy logic. These approaches are highly novel, and we expect that our methods will significantly improve upon current SRGM techniques. Our contributions will benefit the Canadian intelligent systems industry by providing new theoretical guidance and a cohort of HQP trained in intelligent systems.