Researchers · Interviewed at NeurIPS 2025
ISTA

NeurIPS 2025 · December 26, 2025 · 15 min
In this #NeurIPS 2025 highlight, Dingling Yao (ISTA Austria) presents a compelling framework for evaluating causal representation learning using purely observational data - with a fascinating real-world case study from ecology 🐜
Key insights: • High-performing representations can still fail causal inference • Causal representation learning is a measurement problem • T-MEX score allows researchers to evaluate causal representations without randomized interventions