Applying, developing and evaluating Bayesian Network structure learning algorithms to complex real-world datasets .

机器学习技术一直在不断进步，以从数据中学习因果模型(贝叶斯网络)，以便理解复杂的现实世界系统并预测干预措施的影响。然而，大部分进展是在学习和评估合成模型方面取得的，而许多现实世界的方面，如数据缺失或噪声、系统的动态演变和不可测量的变量，相对被忽视。此外，人们较少关注将机器学习与专家知识和实验干预相结合，以及解释机器学习算法为什么会产生它们所做的模型。这项研究将集中于解决这些问题，以便产生更好和更可解释的真实世界系统的因果模型，例如，在健康、社会和环境领域。

There has been continued advances of machine learning techniques to learn causal models (Bayesian Networks) from data in order to understand complex real world systems and predict the effect of interventions in them. However much of the progress has been on learning and evaluating synthetic models, with many real-world aspects such as missing or noisy data, dynamic evolution of the system and unmeasured variables being relatively neglected. Moreover, there has been less focus on integrating machine learning with expert knowledge and experimental interventions, as well as explaining why the machine learning algorithms produce the models that they do. This research will focus on addressing these issues in order to produce better and more explainable causal models of real-world systems in, for example, the health, social and environmental domains.