Zero-shot World Models Are Developmentally Efficient Learners

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Aw, Khai Loong, Kotar, Klemen, Lee, Wanhee, Kim, Seungwoo, Jedoui, Khaled, Venkatesh, Rahul, Chen, Lilian Naing, Frank, Michael C., Yamins, Daniel L. K.
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866908957087367168
author Aw, Khai Loong
Kotar, Klemen
Lee, Wanhee
Kim, Seungwoo
Jedoui, Khaled
Venkatesh, Rahul
Chen, Lilian Naing
Frank, Michael C.
Yamins, Daniel L. K.
author_facet Aw, Khai Loong
Kotar, Klemen
Lee, Wanhee
Kim, Seungwoo
Jedoui, Khaled
Venkatesh, Rahul
Chen, Lilian Naing
Frank, Michael C.
Yamins, Daniel L. K.
contents Young children demonstrate early abilities to understand their physical world, estimating depth, motion, object coherence, interactions, and many other aspects of physical scene understanding. Children are both data-efficient and flexible cognitive systems, creating competence despite extremely limited training data, while generalizing to myriad untrained tasks -- a major challenge even for today's best AI systems. Here we introduce a novel computational hypothesis for these abilities, the Zero-shot Visual World Model (ZWM). ZWM is based on three principles: a sparse temporally-factored predictor that decouples appearance from dynamics; zero-shot estimation through approximate causal inference; and composition of inferences to build more complex abilities. We show that ZWM can be learned from the first-person experience of a single child, rapidly generating competence across multiple physical understanding benchmarks. It also broadly recapitulates behavioral signatures of child development and builds brain-like internal representations. Our work presents a blueprint for efficient and flexible learning from human-scale data, advancing both a computational account for children's early physical understanding and a path toward data-efficient AI systems.
format Preprint
id arxiv_https___arxiv_org_abs_2604_10333
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Zero-shot World Models Are Developmentally Efficient Learners
Aw, Khai Loong
Kotar, Klemen
Lee, Wanhee
Kim, Seungwoo
Jedoui, Khaled
Venkatesh, Rahul
Chen, Lilian Naing
Frank, Michael C.
Yamins, Daniel L. K.
Artificial Intelligence
Computer Vision and Pattern Recognition
Young children demonstrate early abilities to understand their physical world, estimating depth, motion, object coherence, interactions, and many other aspects of physical scene understanding. Children are both data-efficient and flexible cognitive systems, creating competence despite extremely limited training data, while generalizing to myriad untrained tasks -- a major challenge even for today's best AI systems. Here we introduce a novel computational hypothesis for these abilities, the Zero-shot Visual World Model (ZWM). ZWM is based on three principles: a sparse temporally-factored predictor that decouples appearance from dynamics; zero-shot estimation through approximate causal inference; and composition of inferences to build more complex abilities. We show that ZWM can be learned from the first-person experience of a single child, rapidly generating competence across multiple physical understanding benchmarks. It also broadly recapitulates behavioral signatures of child development and builds brain-like internal representations. Our work presents a blueprint for efficient and flexible learning from human-scale data, advancing both a computational account for children's early physical understanding and a path toward data-efficient AI systems.
title Zero-shot World Models Are Developmentally Efficient Learners
topic Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.10333