• Building compositional tasks w...

Treffer: Building compositional tasks with shared neural subspaces.

Title:
Building compositional tasks with shared neural subspaces.
Authors:
Tafazoli S; Princeton Neuroscience Institute, Princeton University, Princeton, NJ, USA. tafazoli@princeton.edu., Bouchacourt FM; Princeton Neuroscience Institute, Princeton University, Princeton, NJ, USA., Ardalan A; Princeton Neuroscience Institute, Princeton University, Princeton, NJ, USA., Markov NT; Princeton Neuroscience Institute, Princeton University, Princeton, NJ, USA., Uchimura M; Princeton Neuroscience Institute, Princeton University, Princeton, NJ, USA., Mattar MG; Department of Psychology, New York University, New York, NY, USA., Daw ND; Princeton Neuroscience Institute, Princeton University, Princeton, NJ, USA.; Department of Psychology, Princeton University, Princeton, NJ, USA., Buschman TJ; Princeton Neuroscience Institute, Princeton University, Princeton, NJ, USA. tbuschma@princeton.edu.; Department of Psychology, Princeton University, Princeton, NJ, USA. tbuschma@princeton.edu.
Source:
Nature [Nature] 2026 Feb; Vol. 650 (8100), pp. 164-172. Date of Electronic Publication: 2025 Nov 26.
Publication Type:
Journal Article
Language:
English
Journal Info:
Publisher: Nature Publishing Group Country of Publication: England NLM ID: 0410462 Publication Model: Print-Electronic Cited Medium: Internet ISSN: 1476-4687 (Electronic) Linking ISSN: 00280836 NLM ISO Abbreviation: Nature Subsets: MEDLINE
Imprint Name(s):
Publication: Basingstoke : Nature Publishing Group
Original Publication: London, Macmillan Journals ltd.
MeSH Terms:
Brain*/physiology , Cognition*/physiology , Macaca mulatta*/physiology , Task Performance and Analysis*, Learning/physiology ; Neurons/physiology ; Psychomotor Performance/physiology ; Animals ; Male ; Models, Neurological ; Neural Networks, Computer
Comments:
Update of: bioRxiv. 2024 Mar 22:2024.01.31.578263. doi: 10.1101/2024.01.31.578263.. (PMID: 38352540)
References:
Nat Hum Behav. 2026 Jan;10(1):111-125. (PMID: 41168425)
Science. 2015 Jun 19;348(6241):1352-5. (PMID: 26089513)
Curr Opin Neurobiol. 2023 Dec;83:102759. (PMID: 37708653)
Neuron. 2023 Dec 6;111(23):3885-3899.e6. (PMID: 37725981)
Proc Natl Acad Sci U S A. 2017 Mar 28;114(13):3521-3526. (PMID: 28292907)
Proc Natl Acad Sci U S A. 2022 Nov;119(44):e2123432119. (PMID: 36279437)
Nat Commun. 2020 Jan 7;11(1):46. (PMID: 31911628)
Trends Cogn Sci. 2019 May;23(5):408-422. (PMID: 31003893)
Trends Cogn Sci. 2022 Jun;26(6):484-498. (PMID: 35469725)
Proc Mach Learn Res. 2017;70:3987-3995. (PMID: 31909397)
Cereb Cortex. 2012 Jun;22(6):1237-46. (PMID: 21817092)
J Neurosci. 2003 Jun 15;23(12):5235-46. (PMID: 12832548)
Nat Commun. 2025 Apr 9;16(1):3359. (PMID: 40204762)
Annu Rev Neurosci. 2008;31:219-45. (PMID: 18558854)
Cell. 2020 Nov 12;183(4):954-967.e21. (PMID: 33058757)
Nature. 2021 Dec;600(7889):489-493. (PMID: 34819674)
Nat Neurosci. 2023 May;26(5):879-890. (PMID: 37024575)
Nat Neurosci. 2019 Oct;22(10):1677-1686. (PMID: 31551604)
Behav Brain Sci. 2017 Jan;40:e253. (PMID: 27881212)
Nat Neurosci. 2024 May;27(5):988-999. (PMID: 38499855)
Nature. 2026 Feb;650(8100):164-172. (PMID: 41299181)
Nature. 2023 Nov;623(7985):115-121. (PMID: 37880371)
J Neurosci. 2017 Jul 19;37(29):6995-7007. (PMID: 28634307)
Nat Commun. 2020 Aug 13;11(1):4069. (PMID: 32792531)
Neuron. 2022 Aug 3;110(15):2503-2511.e3. (PMID: 35700735)
Nature. 2013 Nov 7;503(7474):78-84. (PMID: 24201281)
J Neurosci. 2013 Feb 27;33(9):3844-56. (PMID: 23447596)
Nat Neurosci. 2019 Feb;22(2):297-306. (PMID: 30643294)
Nature. 2021 Apr;592(7855):601-605. (PMID: 33790467)
J Cogn Neurosci. 2018 Aug;30(8):1197-1208. (PMID: 29694261)
Nat Neurosci. 2024 Jul;27(7):1349-1363. (PMID: 38982201)
Nature. 2001 Jun 21;411(6840):953-6. (PMID: 11418860)
Nat Commun. 2016 Oct 27;7:13239. (PMID: 27807345)
Trends Cogn Sci. 2020 Dec;24(12):1028-1040. (PMID: 33158755)
Science. 2011 Mar 11;331(6022):1279-85. (PMID: 21393536)
Nat Neurosci. 2023 Jan;26(1):140-149. (PMID: 36550292)
Nature. 2025 Mar;639(8054):421-429. (PMID: 39608399)
Nature. 2025 Jan;637(8046):663-672. (PMID: 39537930)
Neuron. 2012 Nov 21;76(4):838-846. (PMID: 23177967)
Nature. 2013 May 30;497(7451):585-90. (PMID: 23685452)
Nat Neurosci. 2022 Oct;25(10):1314-1326. (PMID: 36171429)
J Neurosci Methods. 2007 Aug 15;164(1):177-90. (PMID: 17517438)
Nat Mach Intell. 2022;4(12):1185-1197. (PMID: 36567959)
Annu Rev Psychol. 2020 Jan 4;71:273-303. (PMID: 31550985)
Annu Rev Neurosci. 2001;24:167-202. (PMID: 11283309)
Nat Commun. 2023 Oct 27;14(1):6837. (PMID: 37884507)
IEEE Trans Pattern Anal Mach Intell. 2018 Dec;40(12):2935-2947. (PMID: 29990101)
Nat Neurosci. 2025 Mar;28(3):665-675. (PMID: 39930096)
Nat Neurosci. 2023 Nov;26(11):1953-1959. (PMID: 37828227)
J Neurosci. 2011 Jul 27;31(30):10787-802. (PMID: 21795531)
Nat Neurosci. 2024 Jun;27(6):1167-1175. (PMID: 38684894)
Front Hum Neurosci. 2011 Nov 21;5:142. (PMID: 22125519)
Neuron. 2022 Apr 6;110(7):1258-1270.e11. (PMID: 35085492)
Proc Natl Acad Sci U S A. 2024 Apr 2;121(14):e2318521121. (PMID: 38551832)
Elife. 2022 Nov 14;11:. (PMID: 36374181)
Cell. 2024 Mar 14;187(6):1476-1489.e21. (PMID: 38401541)
Grant Information:
R01 MH129492 United States MH NIMH NIH HHS; T32 MH065214 United States MH NIMH NIH HHS
Entry Date(s):
Date Created: 20251126 Date Completed: 20260204 Latest Revision: 20260305
Update Code:
20260305
PubMed Central ID:
PMC12872450
DOI:
10.1038/s41586-025-09805-2
PMID:
41299181
Database:
MEDLINE

Weitere Informationen

Cognition is highly flexible-we perform many different tasks<sup>1</sup> and continually adapt our behaviour to changing demands<sup>2,3</sup>. Artificial neural networks trained to perform multiple tasks will reuse representations<sup>4</sup> and computational components<sup>5</sup> across tasks. By composing tasks from these subcomponents, an agent can flexibly switch between tasks and rapidly learn new tasks<sup>6,7</sup>. Yet, whether such compositionality is found in the brain is unclear. Here we show the same subspaces of neural activity represent task-relevant information across multiple tasks, with each task flexibly engaging these subspaces in a task-specific manner. We trained monkeys to switch between three compositionally related tasks. In neural recordings, we found that task-relevant information about stimulus features and motor actions were represented in subspaces of neural activity that were shared across tasks. When monkeys performed a task, neural representations in the relevant shared sensory subspace were transformed to the relevant shared motor subspace. Monkeys adapted to changes in the task by iteratively updating their internal belief about the current task and then, based on this belief, flexibly engaging the shared sensory and motor subspaces relevant to the task. In summary, our findings suggest that the brain can flexibly perform multiple tasks by compositionally combining task-relevant neural representations.
(© 2025. The Author(s).)

Competing interests: The authors declare no competing interests.