Introductory school, focus on computer science

Topics

Bayesian modeling and Machine Learning:

We will offer an introduction to probabilistic modeling and machine learning, starting from fundamental concepts such as Gaussian random variables and posterior distributions, and extending to advanced topics such as latent variable models and Gaussian Processes. We will explore multivariate Gaussians through a geometric perspective that will lead to an understanding of metrics and inner products in general vector spaces that find applications in machine learning. The concept of generalization will be discussed, including methods for its estimation through techniques like cross-validation.

Applications of Machine Learning in Astronomy:

Following on from the more fundamental introductions, specialists working in the field will present a number of applications of modern machine learning to help solve astrophysical problems.

Connected Morphological Image Processing in Astronomy:

Connected Morphological filters have seen rapid development theory, algorithms, and applications in many fields of computer vision, and have found use in many areas of science, from the subatomic to cosmic scales. In astronomy, the most notable applications in astronomy are in object detection and pattern analysis. A great strength of these methods is that they are scalable to very large data sizes, and offer the option of incorporating explainable AI tools and statistical tests into their frameworks. In the past decade, parallel and distributed algorithms capable of handling giga- and tera-scale data sets are available. In this tutorial we will give an overview of existing tools and algorithms, and new developments in multi-band object detection, dealing with huge data cubes from e.g. LOFAR, exploratory data analysis, and combining these methods with machine-learning tools. A hands-on practical on real data will be provided. 

Deep learning methods for computer vision:

Since about 2012, deep learning has taken the field of computer vision by storm, surpassing the state of the art for most of its tasks, ie object detection and segmentation, image restoration (denoising, deblurring, super-resolution), image reconstruction, 3D field estimation, and much more. Applications in astronomy are numerous. Deep learning relies on the ability of neural network to experimentally learn high quality functions of any order, particularly those associated with signal and image processing. In this short course, we will introduce the basic elements of neural network architectures, the main applications, how to perform effective training, how to deal with limited data and annotations, an introduction to generative methods and also the limitation and explanability of neural networks.

Project Work:

A number of hours will be set aside for practical project work, in which participants of the school will work together in small groups on research projects proposed and supervised by the lecturers on the course. We aim to match students' interests to the projects offered, and a mix of interets and expertise among group members should lead to interesting discussions and learning experiences. At the end of the school, the groups will present their results.

How to write good code:

Two hours tutorial/masterclass on best practice developing good code, primarily in Python, and with applications in ML.

Dealing with Impostor Syndrome:

One-hour session on what impostor syndrome is, and how to deal with it.

Time Management:

One-hour tutorial on best practice in time management. How can we deal with multiple demands on our limited time, while maintaing a healthy work-life balance?