Towards an Open CommonSense Knowledge Base
Abstract
Over the last decade, we have made significant advances in big-data systems using approaches such as deep learning and convolutional neural networks. However, one area where our current systems still lack is reasoning with common sense knowledge. We believe learning common-sense requires integration across multiple sources and multiple modalities of data: for example, it requires using text, visual data and even physical interactions. Similarly, common sense knowledge can be represented in multiple forms: from explicit declarative tuples to implicit procedural knowledge in actions policies. We propose research that will explore different sources and different forms of common sense knowledge to develop a universal open-source common-sense knowledge base. We will also explore cognitive development process for learning common-sense knowledge.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 08, 2019
- Source ID
- W911NF1810019
Entities
People
- Abhinav Gupta
Organizations
- Army Contracting Command
- Defense Advanced Research Projects Agency
- Massachusetts Institute of Technology