As machine learning systems are pervasively deployed in many scientific fields with increasingly sensitive tasks, it has become paramount to develop algorithms that are robust against the numerous sources of uncertainty inherent in those applications including noise in the data, malicious exploitation of vulnerabilities, outliers, variability of the true objective, privacy, and fairness. While current research in machine learning has led to fundamental breakthroughs, there is still a large gap between the theory and the limitations of the existing algorithms used by practitioners in the real world. An important research direction for our group is to lift the current methods out of the sterile lab environment and scale them into the messy real world, by carefully reexamining their limitations, considering more realistic but less perfect conditions, and developing correspondingly robust algorithms.