GPU Cluster for Deep Machine Perception and Interaction

Abstract

This proposal brings together seven faculty members at UT Austin in the areas of machinelearning, vision, robotics, and natural language processing. The common thread of our research ismachine perception and interaction. Our goal is to develop autonomous intelligent systems thatperceive their environment (particularly through vision and language) and fluidly interact withother systems, robots, and human users.With the support of eight ongoing DoD projects among us, the research is proceeding verywell. However, there is one key missing element that undeniably impedes its pace and scope: welack access to a large GPU cluster. Big data and big (many parameter, many layer) networks areessential for capitalizing on the strengths of deep learning, and GPUs are essential to make suchimplementations run in a realistic timeframe. Our current GPU resources are severely limited.That forces us to artificially restrict the scale of our prototypes and experiments, which in turnrisks concealing the full impact of our ideas.Our equipment request is to build a large-scale, cutting edge GPU cluster comprised of a mix ofhigh-end and less expensive GPUs to maximize efficacy for our research. The requested equipmentwould accelerate learning and inference in many cases by orders of magnitude, e.g., opening thepossibility of training a system in days that would now require years using our current resources.This turn-around time is imperative in basic research; not only could we test an idea at scale, butwe could even iteratively improvement it until it really works! In short, the cluster would havedramatic impact on the work of all PIs and their students—not only for the eight current DoDprojects, but also research-related education and new emerging projects over the next 5-10 years.

Document Details

Document Type
DoD Grant Award
Publication Date
Apr 09, 2018
Source ID
FA95501810079

Entities

People

  • Kristen Grauman

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of Texas at Austin

Tags

Fields of Study

  • Computer science

Readers

  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Distributed Systems and Data Platform Development
  • Research Science/Academic Research

Technology Areas

  • AI & ML
  • AI & ML - DoD AI Strategy
  • Autonomy