Multi-Perspective, Spatiotemporal-Coherent, and Multi-Modal Aware Object Detection for Adversarial Robustness

Abstract

In Phase 3, we will fully utilize simulation and our synthetic dataset tools to construct and understand our robust features and representations approach. Synthetic data enables us to independently collect multi-modal data from simulation, while our auto-labeling tools enables us to construct multi-task annotations. Together multi-modal data and multi-task annotations enable implementour robust features and representations. Simulation also enables us to instantiate physically plausible attacks. This enables us to refine our physically plausible threat models and understand our defenses on a more level playing field. In the following sections, we summarize our progress foreach quarter.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
May 30, 2024
Accession Number
AD1229223

Entities

People

  • Cory Cornelius

Tags

Fields of Study

  • Computer science

Readers

  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Software Engineering.
  • Theoretical Analysis.