Using Bayesian Machine Learning to Study Type Ia Supernovae

使用贝叶斯机器学习研究 Ia 型超新星

• 批准号: 2780093
• 金额: --
• 依托单位: University of Cambridge
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2022
• 资助国家: 英国
• 起止时间: 2022 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2780093
• 关键词: Using; Bayesian; Machine; Learning; Study

My PhD project will involve applying cutting-edge machine learning to big data astronomy. I will work on Bayesian models for Type Ia supernovae, supervised by Dr Kaisey Mandel. Type Ia supernovae are standard candles meaning if we have a model for how bright they are, we can determine their distance. These distances can be combined with galaxy redshifts to put constraints on the expansion and age of the Universe using Hubble's Law. The models are perfect for machine learning as they vary in time as well as wavelength, meaning they are high-dimensional. Taking a hierarchical Bayesian approach to this problem allows us to make inferences on other physics models such as dust extinction laws. There are upcoming telescope missions that will provide us with discoveries of unprecedented amounts of supervnoae detections. The Type Ia models need to be scalable to processes the large number of detections, which is why my work will look towards using machine learning methods rather than traditional methods such as Markov Chain Monte Carlo. Further to this, there will be more spectroscopic (high resolution at all wavelengths) supernovae measurements which will provide opportunity to refine our models. It will be my responsibility in the group to find data-efficient machine learning methods to process this high resolution data.

我的博士项目将涉及将尖端机器学习应用于大数据天文学。我将在 Kaisey Mandel 博士的指导下研究 Ia 型超新星的贝叶斯模型。 Ia 型超新星是标准烛光，这意味着如果我们有它们亮度的模型，我们就可以确定它们的距离。这些距离可以与星系红移结合起来，利用哈勃定律对宇宙的膨胀和年龄施加限制。这些模型非常适合机器学习，因为它们随时间和波长而变化，这意味着它们是高维的。采用分层贝叶斯方法解决这个问题使我们能够对其他物理模型（例如尘埃消光定律）进行推断。即将到来的望远镜任务将为我们提供前所未有的超新星探测发现。 Ia 型模型需要可扩展以处理大量检测，这就是为什么我的工作将倾向于使用机器学习方法而不是马尔可夫链蒙特卡罗等传统方法。除此之外，还将有更多的光谱（所有波长的高分辨率）超新星测量，这将为完善我们的模型提供机会。我在小组中的责任是找到数据高效的机器学习方法来处理这些高分辨率数据。