Multi-modal Open World Grounded Learning and Inference (MOWGLI)
Abstract
The scope of this effort is to develop a research-driven general Artificial Intelligence (AI) commonsense service that embodies an innovative, unifying concept called the Multi-modal Open-World Grounded Learning and Inference, or MOWGLI that will harness and synthesize state-of-the-art AI research across representation learning, knowledge graphs, and explainable AI. The MOWGLI system will acquire common-world ‘background’ and context by performing weakly supervised, joint commonsense extraction and computer vision over multi-modal sources such as video and text, combining these extractions with structured knowledge bases (e.g., NELL, YAGO) and concept ontologies (e.g., ConceptNet). The result will be a commonsense knowledge graph that will support a rich set of intuitive everyday phenomena such as abduction, analogy, causality, agency, and physics in a unified reasoning engine.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 13, 2022
- Source ID
- N660011924033
Entities
People
- Pedro Szekely
Organizations
- Defense Advanced Research Projects Agency
- Naval Information Warfare Center Pacific
- University of Southern California