Title: Low-shot Object Learning with Mutual Exclusivity Bias

Date: Friday, December 1, 2023

Time: 10:30AM-12:00PM ET

In-person Location: Coda C1205 Five Points

Zoom link: https://gatech.zoom.us/j/97091549311

Anh Thai

PhD Student in Computer Science

School of Interactive Computing

College of Computing

Georgia Institute of Technology

Committee

Dr. James M. Rehg (advisor), College of Computing, Georgia Institute of Technology; Department of Computer Science and Industrial and Enterprise Systems Engineering, University of Illinois Urbana-Champaign

Dr. Judy Hoffman, College of Computing, Georgia Institute of Technology

Dr. James Hays, College of Computing, Georgia Institute of Technology

Dr. Michael C. Frank, Department of Psychology, Stanford University

Dr. Jia-Bin Huang, Department of Computer Science, University of Maryland, College Park

Summary

Despite rapid development of machine learning techniques that can generalize beyond the distribution of the training data, these models are still far behind the learning pace of young children. In this proposal, we leverage insights from developmental psychology regarding children's learning environment and strategies to apply to machine algorithms. To achieve this goal, we focus on two aspects of children's word and object learning: 3D information and mutual exclusivity bias. We conduct studies on the generalization ability of 3D reconstruction models, identifying key factors that affect this capability. Extending our exploration, we demonstrate that 2D feature representations with strong semantic correspondence matching ability can be effectively employed for 3D object part segmentation. Additionally, we introduce a novel paradigm for low-shot learning, requiring computational models to leverage mutual exclusivity bias to resolve ambiguity in learning signals. Our main goal is to develop a part-based self-supervised learning model that aggregates 3D information from multiple viewpoints. We plan to show that this approach can be applied to address the challenges of low-shot object learning with mutual exclusivity bias setting.