The advent of the Internet gives rise to an exponential growth in data collection, availability and complexity. With it increases our need for more efficient data analysis algorithms. Over super-scale datasets, the only feasible data analysis techniques are iterative linear-time first-order optimization methods.
The computational bottleneck in applying these state-of-the-art iterative methods to machine learning and data analysis is often the so-called "projection step". This project addresses the need to design projection-free optimization algorithms that replace projections by more efficient linear optimization steps. A key contribution of the project is the continual dissemination and transfer of this technology. The open-source software releases will continue to enable large-scale machine learning applications in science and engineering. The broader impact goals of the project, beyond theory and algorithms, include the development of a textbook on efficient optimization techniques in machine learning, as well as the development of a new curriculum focused on preparing students for the scientific and engineering needs in this field.