Flexible Compositional Learning of Structured Visual Concepts

Yanli Zhou; Brenden M. Lake

Flexible Compositional Learning of Structured Visual Concepts

Yanli Zhou
, Brenden M. Lake

Research output: Contribution to conference › Paper › peer-review

Abstract

Humans are highly efficient learners, with the ability to grasp the meaning of a new concept from just a few examples. Unlike popular computer vision systems, humans can flexibly leverage the compositional structure of the visual world, understanding new concepts as combinations of existing concepts. In the current paper, we study how people learn different types of visual compositions, using abstract visual forms with rich relational structure. We find that people can make meaningful compositional generalizations from just a few examples in a variety of scenarios, and we develop a Bayesian program induction model that provides a close fit to the behavioral data. Unlike past work examining special cases of compositionality, our work shows how a single computational approach can account for many distinct types of compositional generalization.

Original language	English (US)
Pages	1146-1152
Number of pages	7
State	Published - 2021
Externally published	Yes
Event	43rd Annual Meeting of the Cognitive Science Society: Comparative Cognition: Animal Minds, CogSci 2021 - Virtual, Online, Austria Duration: Jul 26 2021 → Jul 29 2021

Conference

Conference	43rd Annual Meeting of the Cognitive Science Society: Comparative Cognition: Animal Minds, CogSci 2021
Country/Territory	Austria
City	Virtual, Online
Period	7/26/21 → 7/29/21

All Science Journal Classification (ASJC) codes

Cognitive Neuroscience
Artificial Intelligence
Computer Science Applications
Human-Computer Interaction

Keywords

Bayesian inference
compositionality
concept learning
few-shot learning
visual learning

Cite this

@conference{b9de6829c7134554b4db174bbfc83071,

title = "Flexible Compositional Learning of Structured Visual Concepts",

abstract = "Humans are highly efficient learners, with the ability to grasp the meaning of a new concept from just a few examples. Unlike popular computer vision systems, humans can flexibly leverage the compositional structure of the visual world, understanding new concepts as combinations of existing concepts. In the current paper, we study how people learn different types of visual compositions, using abstract visual forms with rich relational structure. We find that people can make meaningful compositional generalizations from just a few examples in a variety of scenarios, and we develop a Bayesian program induction model that provides a close fit to the behavioral data. Unlike past work examining special cases of compositionality, our work shows how a single computational approach can account for many distinct types of compositional generalization.",

keywords = "Bayesian inference, compositionality, concept learning, few-shot learning, visual learning",

author = "Yanli Zhou and Lake, \{Brenden M.\}",

note = "Publisher Copyright: {\textcopyright} Cognitive Science Society: Comparative Cognition: Animal Minds, CogSci 2021.All rights reserved.; 43rd Annual Meeting of the Cognitive Science Society: Comparative Cognition: Animal Minds, CogSci 2021 ; Conference date: 26-07-2021 Through 29-07-2021",

year = "2021",

language = "English (US)",

pages = "1146--1152",

}

TY - CONF

T1 - Flexible Compositional Learning of Structured Visual Concepts

AU - Zhou, Yanli

AU - Lake, Brenden M.

PY - 2021

Y1 - 2021

N2 - Humans are highly efficient learners, with the ability to grasp the meaning of a new concept from just a few examples. Unlike popular computer vision systems, humans can flexibly leverage the compositional structure of the visual world, understanding new concepts as combinations of existing concepts. In the current paper, we study how people learn different types of visual compositions, using abstract visual forms with rich relational structure. We find that people can make meaningful compositional generalizations from just a few examples in a variety of scenarios, and we develop a Bayesian program induction model that provides a close fit to the behavioral data. Unlike past work examining special cases of compositionality, our work shows how a single computational approach can account for many distinct types of compositional generalization.

AB - Humans are highly efficient learners, with the ability to grasp the meaning of a new concept from just a few examples. Unlike popular computer vision systems, humans can flexibly leverage the compositional structure of the visual world, understanding new concepts as combinations of existing concepts. In the current paper, we study how people learn different types of visual compositions, using abstract visual forms with rich relational structure. We find that people can make meaningful compositional generalizations from just a few examples in a variety of scenarios, and we develop a Bayesian program induction model that provides a close fit to the behavioral data. Unlike past work examining special cases of compositionality, our work shows how a single computational approach can account for many distinct types of compositional generalization.

KW - Bayesian inference

KW - compositionality

KW - concept learning

KW - few-shot learning

KW - visual learning

UR - https://www.scopus.com/pages/publications/85139405611

UR - https://www.scopus.com/pages/publications/85139405611#tab=citedBy

M3 - Paper

AN - SCOPUS:85139405611

SP - 1146

EP - 1152

T2 - 43rd Annual Meeting of the Cognitive Science Society: Comparative Cognition: Animal Minds, CogSci 2021

Y2 - 26 July 2021 through 29 July 2021

ER -

Flexible Compositional Learning of Structured Visual Concepts

Abstract

Conference

All Science Journal Classification (ASJC) codes

Keywords

Other files and links

Fingerprint

Cite this