Selected topics in machine learning and AI applications in science

• Students will learn the basics of machine learning data analysis techniques
• They will be able to apply these techniques to a wide range of data, including different types of scientific observation data
• They will get an overview of how modern ML/AI is used in science

To follow the course basic coding skills (Python) are needed

The course introduces the basic principles of ML in a series of lectures and covers advanced topics in seminars.

These topics include:

Introductory part:
- perceptron, gradient descent and backpropagation;
- Regression and Classification tasks;
- basics of Supervised, Unsupervised and Semi-supervised methods
- data preparation
- Decision trees and Random forest
- Deep Neural Network: Convolution and Convolutional Neural Network; Applications on Image Analysis

Advanced topics:
- Statistical Techniques for Modern ML: Maximum Likelihood, Bayes’ Rule, Maximum A Posteriori (MAP), KL Divergence, Entropy; Applications with Variational Autoencoder (Data Generation)
- Simulation-Based Inference, Attention Mechanism and Transformers
- Generative Adversarial Networks, Diffusion Models, Contrastive Learning
- LLMs and foundation models

• Familiarisation with modern AI algorithms
• Application of these algorithms to real-world data sets
• Ready to apply the techniques learnt this course to your own use case

Report on the final project

• AutoSourceID-Classifier - Star-galaxy classification using a convolutional neural network with spatial information, F. Stoppa, S. Bhattacharyya, G. Zaharijas et al. Published in: Astron.Astrophys. 680 (2023) A109
• Mind the gap: the discrepancy between simulation and reality drives interpretations of the Galactic Center Excess, Sascha Caron , Christopher Eckner, Gabrijela Zaharijas et al. Published in: JCAP 06 (2023) 013
• AutoSourceID-Light - Fast optical source localization via U-Net and Laplacian of Gaussian, Fiorenzo Stoppa, S. Bhattacharyya, G. Zaharijas et al. Published in: Astron.Astrophys. 662 (2022) A109
• Identification of point sources in gamma rays using U-shaped convolutional neural networks and a data challenge, Boris Panes, Christopher Eckner, Gabrijela Zaharijas et al. Published in: Astron.Astrophys. 656 (2021) A62