HBKU AI for Science Symposium

In the first hour of this tutorial, I will give a brief overview of key techniques in physics-based deep learning. I will then dive into state-of-the-art neural operator methods, discussing approaches for both uniform and non-uniform discretizations of physical systems. We will compare the strengths and limitations of different neural approaches and the types of problems they are suited to. The tutorial will cover Fourier and convolutional neural operators for structured data, as well as graph-based neural networks for unstructured domains.

Materials informatics—the application of data science and machine learning to materials research—has become a powerful approach for accelerating materials discovery and design. Unlike traditional machine learning, materials informatics must contend with small, heterogeneous, and expensive datasets, strong physical constraints, and the need for interpretability and uncertainty awareness. This workshop introduces the foundations of materials informatics, explaining why data-driven methods are needed, how they differ from conventional ML workflows, and how they fit within the structure–property–processing–performance paradigm. The workshop then focuses on best practices for turning materials data into actionable knowledge and ultimately into new materials. Topics include data sourcing and cleaning, featurization and representation of materials, model selection and validation, avoiding common pitfalls such as data leakage, and using uncertainty-aware models for decision-making. Through practical examples and case studies—including predictive modeling and generative approaches—participants will learn how to move from models to real materials applications using open, reproducible tools.

In this tutorial, we shall cover machine learning and data science concepts related to biology. We will expose the underlying machine learning and biology principles using a few exercises, each covering a different current area in biology. We will use the TRILL package to download, fine tune and save a protein language model like alphafold. We will also build a classification model that uses tabular features to diagnose high grade cancer. Lastly, we will look at differential expression using gaussian mixture models, to identify genes of interest relevant to cancer. Together, we will cover multiple key areas of machine learning research via language models, graph models and classification models and biology via protein language models, gene expression and hallmarks of cancer.