Robust Estimation

Robust estimation studies how to recover useful parameters, predictions, or system states when measurements may contain outliers, heavy-tailed noise, adversarial contamination, or distribution shifts. Common approaches include median-of-means and other robust aggregation, M-estimators and trimming, filtering, RANSAC-style consensus, and distributionally robust optimization; they are used in regression, mean and covariance estimation, computer vision, and robotics. Researchers analyze finite-sample error and recovery guarantees, and evaluate methods on settings such as contaminated regression data, geometric matching benchmarks, and sensor-fusion tasks. A central tension is limiting the influence of bad data without sacrificing accuracy on clean data, while also keeping methods computationally practical and able to adapt to changing conditions.