Reasoning in autonomous systems via wiring diagrams
Abstract
Machine learning is one of the most prominent approaches in artificial intelligence today. More and more, we are finding ourselves relying on machine learning in tasks of all levels of scale and importance. A bottleneck in this trend, however, is the statistical approach underlying machine-learning algorithms. For these statistical methods to perform well, a large amount of data is needed. In addition, the nature of these algorithms can make it difficult for humans to understand how an algorithm arrives at a decision. In contrast, humans can learn from a small number of examples. Humans also have the ability to explain the process of how they arrive at a decision. Enabling machines to have these traits of human reasoning is a fundamental challenge. In this project, we will tackle this challenge by building a mathematical theory of reasoning and problem solving that is context-independent. The technical foundation of our approach will be category theory, a branch of mathematics that studies the relations among mathematical structures in different fields. We will develop an arithmetic system for wiring diagrams, which are mathematical objects that represent abstract concepts. Calculations in this arithmetic system will represent processes of reasoning and problem solving. Since the reasoning framework we aim to build will be context-independent, it can be used in autonomous systems in any application domain. Systems that utilize our reasoning framework will be able to solve unfamiliar problems via analogical reasoning. Under our framework, the process behind every decision will also be transparent; this will help build trust between humans and machines when they work together as a team.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 06, 2025
- Source ID
- FA95502410268
Entities
People
- Jason Lo
Organizations
- Air Force Office of Scientific Research
- United States Air Force