Statistical inference using machine learning methods

机器学习领域为我们提供了一系列强大的预测方法，其中包括随机森林，提升树，核机器和神经网络等，在各种应用中非常成功。然而，在许多情况下，获得预测并不是唯一的目标，但我们希望了解哪些变量是重要的，以及它们如何影响响应，并量化我们结论的不确定性。虽然这些任务的推理工具是现成的标准统计方法，这些通常依赖于不切实际的建模假设。该项目将开发旨在利用机器学习方法的预测能力的方法，以提供这种推理分析，而不受经典模型的限制。

The field of machine learning has given us an array of powerful prediction methods with random forests, boosted trees, kernel machines and neural networks, among others, being highly successful in a variety of applications. There are however many settings where obtaining predictions is not the only goal, but we would like to understand which variables are important and how they influence the response, and also quantify the uncertainty of our conclusions. Whilst the inferential tools for these tasks are readily available for standard statistical methods, these typically rely on unrealistic modelling assumptions. This project will develop approaches that aim to harness the predictive ability of machine learning methods to provide this sort of inferential analysis without the restrictions of classical models.