MEMY-599
Machine Learning theory and methods

Syllabus

The syllabus covers the following topics organized in the following 5 modules:

Introduction

Introduction and background to the topic

Fundamentals

• Machine Learning: concept, categories, applicability, features and extraction
• Data: data structures, use, storage (FAIR principles), platforms and databases
• Features: importance and extraction, examples, workflows
• Explainability: use cases and critical assessment
• Ensemble learning: boosting, trapping

Methods and algorithms

• Recap: Unsupervised and supervised learning: selected methods (principal component analysis, kmeans, DBSCAN, regression)
• Neural Networks: mathematical build-up, hyperparameters, selected networks (long short-term memory, graph neural networks), autoencoders, physics-informed neural networks
• Generative models: concepts and differences from non-generalized models, examples (variational autoencoders, general adversial networks)
• Hybrid models: concepts, examples
• Large language models: transformers, tokenization

Design of soft materials & (bio)molecules

• Structure: proteins (AlphaFold), molecules, fingerprinting/featurization, selected prediction models
• Properties: embeddings, use cases

Machine Learning and computer simulations

• Potentials: development generations, distinct energy descriptions
• Interactions: short-, medium-, and long-range

Learning Outcomes

Upon successful completion of the course students will be able to

• Gain experience with data and databases and be aware of the fairness in terms of generating, using, and archiving data.
• Become familiar with advanced concepts of Machine Learning with emphasis in Natural Sciences and Engineering.
• Critically assess the applicability, explainability, and generalizability of Machine Learning methods.
• Understand the conceptual differences, strengths & weaknesses of standard with respect to generative learning models.
• Critically compare Machine Learning schemes and algorithms.
• Express a Machine Learning algorithm in form of a pseudocode.
• Express solid-state structures and molecules as embeddings and be able to provide examples and ideas of molecular fingerprints.
• Understand the concept of Machine Learning potentials and explain their applicability.
• Gain a practical hands-on experience with advanced Machine Learning algorithms.

Suggested Bibliography

1. I. Goodfellow, Y. Bengio, A. Courville, Deep Learning (Adaptive Computation and Machine Learning series), The MIT Press, USA (2016)
2. M.P. Deisenroth, A. A. Faisal, C.S. Ong, Mathematics for Machine Learning, Cambridge University Press (2020)
3. K.P. Murphy, Machine Learning a probabilistic perspective, The MIT Press (2012)
4. S. Shalev-Shwartz, S. Ben-David, Understanding Machine Learning, Cambridge University Press (2014)
5. M. Erdmann, J. Glombitza, G. Kasieczka, U. Klemradt, Deep Learning for Physics Research, World Scientific (2021).