Mental bootstrapping enables human-level concept learning in self-supervised deep models
Science Advances · 2026
DOI: 10.1126/sciadv.aea7202 ↗Kurz gefasst
Fachartikel in der Zeitschrift Science Advances, 2026. Verfasst von Lingxiao Yang und 6 weiteren Autor:innen. Beteiligte Einrichtung: School of Systems Science and Engineering, Sun Yat-sen University, Guangzhou 510006, Guangdong, People's Republic of China. Hauptthemen laut MEDLINE: Concept Formation, Supervised Machine Learning, Learning, Deep Learning. Die Arbeit nennt 61 Literaturangaben. Der Volltext ist frei in PubMed Central verfügbar. Gefördert von: National Natural Science Foundation of China, Shanghai Municipal Education Commission, Beijing Natural Science Foundation.
Abstract (Originalsprache)
How agents acquire abstract concepts from sparse, diverse examples—often without explicit supervision—remains a central problem in cognitive science and artificial intelligence. Human studies suggest that this ability depends on mental bootstrapping, the gradual construction of complex concepts from simpler partial structures. Building on this idea, we develop a self-supervised framework that trains models on systematically simplified versions of abstract reasoning tasks containing incomplete but structured concept cues. This algorithm enables models to form internal abstractions under limited resources and later apply them to more complex problems. We evaluate the framework across 12 abstract visual reasoning datasets testing in-distribution concept induction, out-of-distribution generalization, and few-shot learning. To contextualize performance, we also measure human accuracy on the same tasks. Models trained on simplified problems generalize robustly, reaching or even surpassing human-level performance. These findings show that abstract reasoning can emerge from structured simplification and minimal data, offering a computational account of concept learning in humans and machines.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Autor:innen und Institutionen
- School of Systems Science and Engineering, Sun Yat-sen University, Guangzhou 510006, Guangdong, People's Republic of China.
- Peking-Tsinghua Center for Life Sciences, Academy for Advanced Interdisciplinary Studies, Peking University, Beijing 100871, People's Republic of China.
- Brain Health Institute, National Center for Mental Disorders, Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine and School of Psychology, Shanghai 200030, People's Republic of China.
- School of Computer Science and Engineering, Sun Yat-sen University, Guangzhou 510006, Guangdong, People's Republic of China.
- Beijing Key Laboratory of Applied Experimental Psychology, Faculty of Psychology, Beijing Normal University, Beijing 100875, People's Republic of China.
- School of Psychological and Cognitive Sciences, Peking University, Beijing 100871, People's Republic of China.
ORCID-Profile: Lingxiao Yang, Muyang Lyu, Ying-Jie Wang, Xiaohua Xie, Zonglei Zhen, Ru-Yuan Zhang
Förderung
- National Natural Science Foundation of China
- Shanghai Municipal Education Commission
- Beijing Natural Science Foundation
Zitieren
APA
Yang, L., Lyu, M., Wang, Y. J., Xie, X., Zhen, Z., Lai, J. H., & Zhang, R. Y. (2026). Mental bootstrapping enables human-level concept learning in self-supervised deep models. Science Advances. https://doi.org/10.1126/sciadv.aea7202
Vancouver
Yang L, Lyu M, Wang YJ, Xie X, Zhen Z, Lai JH, et al. Mental bootstrapping enables human-level concept learning in self-supervised deep models. Science Advances. 2026. doi:10.1126/sciadv.aea7202
BibTeX
@article{Yang2026,
author = {Yang, Lingxiao and Lyu, Muyang and Wang, Ying-Jie and Xie, Xiaohua and Zhen, Zonglei and Lai, Jian-Huang and Zhang, Ru-Yuan},
title = {Mental bootstrapping enables human-level concept learning in self-supervised deep models},
journal = {Science Advances},
year = {2026},
doi = {10.1126/sciadv.aea7202}
}
EUVIMED