SPARC: A Human-in-the-Loop Framework for Learning and Explaining Spatial Concepts

In this article, we introduce a novel framework for learning spatial concepts within a human-in-the-loop (HITL) context, highlighting the critical role of explainability in AI systems. By incorporating human feedback, the approach enhances the learning process, making it particularly suitable for ap...

Full description

Saved in:
Bibliographic Details
Main Authors: Brendan Young, Derek T. Anderson, James Keller, Frederick Petry, Chris J. Michael
Format: Article
Language:English
Published: MDPI AG 2025-03-01
Series:Information
Subjects:
Online Access:https://www.mdpi.com/2078-2489/16/4/252
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In this article, we introduce a novel framework for learning spatial concepts within a human-in-the-loop (HITL) context, highlighting the critical role of explainability in AI systems. By incorporating human feedback, the approach enhances the learning process, making it particularly suitable for applications where user trust and interpretability are essential, such as AiTR. Namely, we introduce a new parametric similarity measure for spatial relations expressed as histograms of forces (HoFs). Next, a spatial concept is represented as a spatially attributed graph and HoF bundles. Last, a process is outlined for utilizing this structure to make decisions and learn from human feedback. The framework’s robustness is demonstrated through examples with diverse user types, showcasing how varying feedback strategies influence learning efficiency, accuracy, and ability to tailor the system to a particular user. Overall, this framework represents a promising step toward human-centered AI systems capable of understanding complex spatial relationships while offering transparent insights into their reasoning processes.
ISSN:2078-2489