Burak Varıcı: Causal Representation Learning скачать в хорошем качестве

Burak Varıcı: Causal Representation Learning

The talk given by Burak Varıcı in KUIS AI Talks on October 21, 2024 Title: Causal Representation Learning Abstract: Representation learning has shown great success at learning low-dimensional, empirically useful representations of high-dimensional data in an unsupervised manner. Despite this success, standard approaches are limited to association-level insights and are inadequate for strong generalization — transfer of the learned representations to new problems. As a key step toward strong generalization and more principled representation learning, causal representation learning (CRL) has emerged as a cutting-edge field that merges the strengths of statistical inference, machine learning, and causal inference. Its objective is to estimate the ground truth, identifiable latent representations and rich structures that model the interactions among the variables in the latent space. In this talk, we will introduce the core concepts and motivations behind combining representation learning with causal inference, including primary objectives and the theoretical challenges. We will then show how interventional or multi-environment data can help establish identifiability proofs and lead to scalable algorithms across parametric and nonparametric settings. A central theme is the connection between CRL and score functions (which underlie diffusion models), yielding practical estimators that exploit modern generative tooling. We will also highlight real-world application opportunities, including a demonstration of CRL on semi-synthetic robotics data. Short Bio: Burak Varıcı is a postdoctoral researcher in the Machine Learning Department at Carnegie Mellon University. Dr. Varıcı received his Ph.D. in Electrical Engineering from Rensselaer Polytechnic Institute in 2024. His research develops theoretical foundations at the intersection of causality and machine learning, with an emphasis on leveraging interventional data and shared causal mechanisms for representation learning, sequential intervention design, and causal structure learning. He is also broadly interested in identifiable representation learning, including self-supervised approaches. He was a recipient of the IBM AI Horizons Fellowship during his Ph.D. Previously, he earned his B.S. in Electrical and Electronics Engineering from Boğaziçi University in 2018.