Foundations of Scalable Statistical Learning

Abstract

Massive data collection by existing array of intelligence sources and services, together with the proliferation of Internet of Things and other Big Data technologies have presented the Defense enterprise with unprecedented challenges and opportunities. On the one hand, the availability of such massive data sets holds the promise of a better functionality across all services, leading to better understanding of complex operations and ultimately to better decisions. On the other hand, overabundance of battlefield data which often is incomplete, contains errors, is sequential in nature, and needs to be processed in real-time to give unambiguous tracks and target state vectors and to provide actionable information to the warfighter, has lead to the problem of data deluge. As various assets gather data at unprecedented scales, filtering and processing this interconnected data into a form that can be used in real-time by the decision maker has become a major bottleneck. Larger datasets not only call for faster processing methods, but also lead us to completely rethink the way data should be modeled. To address the above challenges, we proposed a bold, multidisciplinary research program in three major thrusts: The first thrust lead to a transformative theory for developing limits and statistical-computational tradeoffs for learning rich graphical models which provide a powerful and general framework to capture the dependencies among entities generating large-dimensional data sets. In addition, we developed a theory for causal and time-dependent inference using graphical models. In the second thrust, we developed a theory for learning with sequential and incomplete data and also focused on the empirical risk minimization problem, which is a general formulation in machine learning for estimating model parameters by minimizing an objective function given as a loss function averaged over data (i.e., a large sum of component functions). We developed online learning tools for contextual monitoring, as well as incremental, distributed, and second order methods for Empirical Risk Minimization. We also developed new insights for geometric optimization and optimization over manifolds. In the third thrust, we used matrix completion as a test case to study statistical and computational trade offs

Document Details

Document Type
DoD Grant Award
Publication Date
Dec 04, 2018
Source ID
W911NF1610551

Entities

People

  • Ali Jadbabaie

Organizations

  • Army Contracting Command
  • Defense Advanced Research Projects Agency
  • Massachusetts Institute of Technology

Tags

Fields of Study

  • Computer science

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Neural Network Machine Learning.
  • Systems Analysis and Design

Technology Areas

  • 5G
  • 5G - Internet of Things
  • AI & ML
  • AI & ML - DoD AI Strategy
  • AI & ML - Machine Learning Algorithms