OPTiMaLOptimization for Machine Learning: from Robustness to Regularization
Abstract
This project will develop and integrate the latest optimization and statistical advances into a new generation of resource-efficient algorithms for large-scale machine learning. State-of-the-art machine learning methods provide impressive results, opening new perspectives for science, technology, and society. However, they rely on massive computational resources to process huge manually annotated data-sets. The corresponding costs in terms of energy consumption and human efforts are not sustainable. This project builds on the idea that improving efficiency is a key to scale the ambitions and applicability of machine learning. Achieving efficiency requires overcoming the traditional boundaries between statistics and computations, to develop new theory and algorithms. Within a multidisciplinary approach, we will establish a new regularization theory of efficient machine learning. We will develop statistical models that incorporate budgeted computations, and numerical solutions with resources tailored to the statistically accuracy allowed by the data. Theoretical advances will provide the foundations for novel and sound algorithmic solutions. The new algorithms developed in the project will contribute to boost the possibilities of Artificial Intelligence, modeling, and decision making in a world of data with ever-increasing size and complexity.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 24, 2019
- Source ID
- FA95501817009
Entities
People
- Rosasco Lorenzo
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Genoa