Outlier-Robust Diffusion Solvers for Inverse Problems
The episode discusses a paper titled "Outlier-Robust Diffusion Solvers for Inverse Problems." The hosts explore how this research modifies diffusion solvers to handle bad data by down-weighting outliers. They discuss the shift from fitting all data points to prioritizing reliability, the potential for standardized robustness across domains, and the implications for deploying reliable AI in high-stakes environments.