Predictive machine learning (ML), such as deep learning, excels at pattern recognition tasks like self-driving cars, protein structure prediction, or large language models. However, these models cannot answer what-if questions such as: “What if we gave a promotional discount, would the revenue increase?” or “What if we gave a different drug to a patient, would they recover faster?” Causal ML provides tools to answer these interventional and counterfactual questions from data. This talk offers a gentle introduction to causal ML, highlights real-world use cases, and walks through a hands-on example using the open-source Python library pgmpy.
Causal inference methods are designed to answer interventional and counterfactual what-if questions. Unlike predictive ML models that rely on correlations, causal methods aim to model the underlying mechanism of the data-generating process. This mechanistic understanding is crucial for applications such as identifying effective interventions, making out-of-distribution predictions, and building explainable AI systems. This talk provides a beginner-friendly overview of causal ML, highlighting scenarios where predictive ML falls short, and where causal inference can provide answers. The main goal of this talk is to help the audience understand the kinds of applications that causal modelling can be applied to.
Understand the limitations of predictive ML: Learn why traditional machine learning models (like deep learning) fail to answer "what-if" or counterfactual questions.
Introduction to Causal Machine Learning (Causal ML): Get a beginner-friendly overview of causal ML, its concepts, and how it differs from predictive ML.
Grasp the importance of causal inference: Understand how causal inference helps uncover the mechanisms behind data instead of just relying on correlation.
Learn about: pgmpy which is a Python library for causal and probabilistic modeling using Bayesian Networks and related models