AI in Science and Engineering Summer Academy 2023

Explainable AI

Explainable AI (XAI) has recently become a major field of research to address the need for better interpretability of data-driven models. The opaqueness of elaborate black-box learning models has impeded trust and acceptance, particularly in industries where high-stakes decision-making is common. This workshop will focus on equipping users with a set of tools to help them explain model behavior, with particular emphasis placed on techniques to obtain model-agnostic explanations for predictions.

This workshop will cover a range of topics related to XAI methodology, including inherently interpretable modeling, anomaly detection and interpretation, counterfactual explanations, model-specific explanations (e.g. TreeSHAP), and model-agnostic explanation techniques (e.g. LIME, SHAP). Attendees of this tutorial will gain hands-on experience applying XAI algorithms and visualizations to various datasets, imparting practical knowledge to enable and encourage users to implement these tools for their own needs.

Natural Language Processing

Natural Language Processing (NLP) is an Artificial Intelligence branch focused on enabling computers to understand, interpret, and generate human language. This field has garnered increasing attention due to recent advances in technology, including accessible chatbots such as ChatGPT. NLP has a broad range of scientific applications, such as language translation, sentiment analysis, speech recognition, information retrieval, and text summarization.

This workshop aims to equip participants with foundational NLP knowledge to enable coding for textual data analysis. It will primarily focus on practical applications of NLP models, utilizing big data sets scraped from social media websites such as Twitter. Throughout the workshop, participants will learn a range of methodologies and techniques that underlie NLP model development, culminating in sentiment analysis of data sets containing public attitudes and opinions on current environmental issues.

Causal Inference

Causal Inference (CI) is the area of study that involves identifying and measuring cause-effect relationships among different variables. In order to do this, CI employs a statistical framework, with various inference methods and their respective assumptions with which it can disentangle correlation from causation. Given the interdisciplinary roots of CI, this workshop will cover methods developed from statistics, computer science, biostatistics, economics, etc., all of which are applicable in various fields of science. The workshop will cover a broad range of topics, including counterfactual inference, randomized experiments, natural experiments, quasi-experimental methods, and causal graphs. Participants will learn how to evaluate the validity of causal inference claims and how to address issues such as selection bias, confounding variables, and measurement error. By the end of this workshop, participants will be able to grasp the main principles of CI and will be able to apply those methods to real-world problems in science and engineering.

Probabilistic Programming

Bayesian methods are a set of statistical techniques that use prior knowledge and probability theory to update beliefs and make predictions about uncertain events. These techniques are increasingly important in scientific computing where uncertainties and incomplete information play a role. This mini-tutorial will walk participants through the basic probabilistic concepts to give the understanding of how parameter estimation problems can be recast in the form of Bayesian inference, as well as equip them with hands-on tools for tackling such problems.

We will explain two main paradigms for posterior estimation: exact Bayesian inference and approximate techniques. We will also introduce probabilistic programming languages are in the modern landscape, demonstrating abilities of one of such easy-to-use software packages.