Francesco Locatello: Causal effects in the eyes of the beholder
THIS TALK HAS BEEN POSTPONED
Date: POSTPONED TO SEPTEMBER
Location: Medical University of Vienna, Anna Spiegel Research Building, Seminar Room Level 3
Speaker: Francesco Locatello
Title: Causal effects in the eyes of the beholder
Abstract: Deciphering raw, high-dimensional, and temporal observations into causal knowledge is a key component of the scientific discovery process and a longstanding challenge for AI. Across scientific disciplines, the data that can be recorded do not directly expose causal variables, which often remain latent and only indirectly measured. In this talk, I present our recent work on accurate causal effect estimation from raw experimental data using deep learning and its interdisciplinary applications. I begin by defining when a predictor constitutes a causally valid proxy of a latent variable, and how deep learning models can process entire experiments to yield correct causal conclusions. I then show how AI models enable "looking at the data first" and discovering treatment effects without supervision using mechanistic interpretability tools alongside statistical testing procedures. I will then show that these methods work even on complex policy problems, discovering actionable sources of heterogeneity in anti-poverty programs. I will conclude with a discussion on identifiability in machine learning, in particular sparse autoencoders, which are the central methodological tool we used in exploratory causal inference.
Bio: Francesco Locatello is a tenure-track assistant professor at the Institute of Science and Technology Austria (ISTA) and an AI resident at the Chan Zuckerberg Initiative. Before, he was a senior applied scientist at Amazon Web Services, leading the Causal Representation Learning team. He received his PhD from ETH Zürich co-advised by Gunnar Rätsch and Bernhard Schölkopf. His research received several awards, including the ICML 2019 Best Paper award, the Hector Foundation award for outstanding achievements in machine learning from the Heidelberg Academy of Science in 2023, and the Google Research Scholar Award in 2024.
This is part of the CAIM Talks series.