Task representations in neural networks trained to perform many cognitive tasks

Guangyu Robert Yang, Madhura R. Joglekar, H. Francis Song, William T. Newsome, Xiao-Jing Wang

Research output: Contribution to journalArticle

Abstract

The brain has the ability to flexibly perform many tasks, but the underlying mechanism cannot be elucidated in traditional experimental and modeling studies designed for one task at a time. Here, we trained single network models to perform 20 cognitive tasks that depend on working memory, decision making, categorization, and inhibitory control. We found that after training, recurrent units can develop into clusters that are functionally specialized for different cognitive processes, and we introduce a simple yet effective measure to quantify relationships between single-unit neural representations of tasks. Learning often gives rise to compositionality of task representations, a critical feature for cognitive flexibility, whereby one task can be performed by recombining instructions for other tasks. Finally, networks developed mixed task selectivity similar to recorded prefrontal neurons after learning multiple tasks sequentially with a continual-learning technique. This work provides a computational platform to investigate neural representations of many cognitive tasks.

Original languageEnglish (US)
Pages (from-to)297-306
Number of pages10
JournalNature Neuroscience
Volume22
Issue number2
DOIs
StatePublished - Feb 1 2019

Fingerprint

Learning
Aptitude
Short-Term Memory
Decision Making
Neurons
Brain

ASJC Scopus subject areas

  • Neuroscience(all)

Cite this

Task representations in neural networks trained to perform many cognitive tasks. / Yang, Guangyu Robert; Joglekar, Madhura R.; Song, H. Francis; Newsome, William T.; Wang, Xiao-Jing.

In: Nature Neuroscience, Vol. 22, No. 2, 01.02.2019, p. 297-306.

Research output: Contribution to journalArticle

Yang, Guangyu Robert ; Joglekar, Madhura R. ; Song, H. Francis ; Newsome, William T. ; Wang, Xiao-Jing. / Task representations in neural networks trained to perform many cognitive tasks. In: Nature Neuroscience. 2019 ; Vol. 22, No. 2. pp. 297-306.
@article{e99f3c9f85af4527916420ba5e05d769,
title = "Task representations in neural networks trained to perform many cognitive tasks",
abstract = "The brain has the ability to flexibly perform many tasks, but the underlying mechanism cannot be elucidated in traditional experimental and modeling studies designed for one task at a time. Here, we trained single network models to perform 20 cognitive tasks that depend on working memory, decision making, categorization, and inhibitory control. We found that after training, recurrent units can develop into clusters that are functionally specialized for different cognitive processes, and we introduce a simple yet effective measure to quantify relationships between single-unit neural representations of tasks. Learning often gives rise to compositionality of task representations, a critical feature for cognitive flexibility, whereby one task can be performed by recombining instructions for other tasks. Finally, networks developed mixed task selectivity similar to recorded prefrontal neurons after learning multiple tasks sequentially with a continual-learning technique. This work provides a computational platform to investigate neural representations of many cognitive tasks.",
author = "Yang, {Guangyu Robert} and Joglekar, {Madhura R.} and Song, {H. Francis} and Newsome, {William T.} and Xiao-Jing Wang",
year = "2019",
month = "2",
day = "1",
doi = "10.1038/s41593-018-0310-2",
language = "English (US)",
volume = "22",
pages = "297--306",
journal = "Nature Neuroscience",
issn = "1097-6256",
publisher = "Nature Publishing Group",
number = "2",

}

TY - JOUR

T1 - Task representations in neural networks trained to perform many cognitive tasks

AU - Yang, Guangyu Robert

AU - Joglekar, Madhura R.

AU - Song, H. Francis

AU - Newsome, William T.

AU - Wang, Xiao-Jing

PY - 2019/2/1

Y1 - 2019/2/1

N2 - The brain has the ability to flexibly perform many tasks, but the underlying mechanism cannot be elucidated in traditional experimental and modeling studies designed for one task at a time. Here, we trained single network models to perform 20 cognitive tasks that depend on working memory, decision making, categorization, and inhibitory control. We found that after training, recurrent units can develop into clusters that are functionally specialized for different cognitive processes, and we introduce a simple yet effective measure to quantify relationships between single-unit neural representations of tasks. Learning often gives rise to compositionality of task representations, a critical feature for cognitive flexibility, whereby one task can be performed by recombining instructions for other tasks. Finally, networks developed mixed task selectivity similar to recorded prefrontal neurons after learning multiple tasks sequentially with a continual-learning technique. This work provides a computational platform to investigate neural representations of many cognitive tasks.

AB - The brain has the ability to flexibly perform many tasks, but the underlying mechanism cannot be elucidated in traditional experimental and modeling studies designed for one task at a time. Here, we trained single network models to perform 20 cognitive tasks that depend on working memory, decision making, categorization, and inhibitory control. We found that after training, recurrent units can develop into clusters that are functionally specialized for different cognitive processes, and we introduce a simple yet effective measure to quantify relationships between single-unit neural representations of tasks. Learning often gives rise to compositionality of task representations, a critical feature for cognitive flexibility, whereby one task can be performed by recombining instructions for other tasks. Finally, networks developed mixed task selectivity similar to recorded prefrontal neurons after learning multiple tasks sequentially with a continual-learning technique. This work provides a computational platform to investigate neural representations of many cognitive tasks.

UR - http://www.scopus.com/inward/record.url?scp=85060098347&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85060098347&partnerID=8YFLogxK

U2 - 10.1038/s41593-018-0310-2

DO - 10.1038/s41593-018-0310-2

M3 - Article

VL - 22

SP - 297

EP - 306

JO - Nature Neuroscience

JF - Nature Neuroscience

SN - 1097-6256

IS - 2

ER -