Machine Learning and Data Mining have gained importance across many areas of science and other industries. A critical challenge is to design learning algorithms that work on real-world data sets, which are often noisy and almost never perfectly fit any particular model. Truly robust and noise-tolerant learning algorithms are necessary for improved predictions, compression, medical diagnoses, and automation. The potential applications of such algorithms are as wide-spread and varied as those of the field of Statistics.
More specifically, the project entails designing machine learning algorithms that are provably robust: ones that optimally fit noisy data. The project has several novel algorithmic and analytical ideas for designing provably noise-tolerant learning algorithms. As with all such ?agnostic? learning algorithms, these algorithms are also computationally efficient. Partly due to the wide variety of applications, theoretically-inspired algorithms have long been the state-of-the-art for machine learning and data mining. The research results of this project have broader impacts across a number of scientific, medical, and industrial fields. The project?s impact extends to academia through educational efforts, including graduate and undergraduate training, curriculum development, novel educational endeavors, and seminars.