Computation noise promotes zero-shot adaptation to uncertainty during decision-making in artificial neural networks.

Publisher: American Association for the Advancement of Science Country of Publication: United States NLM ID: 101653440 Publication Model: Print-Electronic Cited Medium: Internet ISSN: 2375-2548 (Electronic) Linking ISSN: 23752548 NLM ISO Abbreviation: Sci Adv Subsets: MEDLINE

Original Publication: Washington, DC : American Association for the Advancement of Science, [2015]-

Neural Networks, Computer* ; Decision Making*/physiology; Uncertainty ; Humans ; Cognition/physiology ; Adult ; Male

Random noise in information processing systems is widely seen as detrimental to function. But despite the large trial-to-trial variability of neural activity, humans show a remarkable adaptability to conditions with uncertainty during goal-directed behavior. The origin of this cognitive ability, constitutive of general intelligence, remains elusive. Here, we show that moderate levels of computation noise in artificial neural networks promote zero-shot generalization for decision-making under uncertainty. Unlike networks featuring noise-free computations, but like human participants tested on similar decision problems (ranging from probabilistic reasoning to reversal learning), noisy networks exhibit behavioral hallmarks of optimal inference in uncertain conditions entirely unseen during training. Computation noise enables this cognitive ability jointly through "structural" regularization of network weights during training and "functional" regularization by shaping the stochastic dynamics of network activity after training. Together, these findings indicate that human cognition may ride on neural variability to support adaptive decisions under uncertainty without extensive experience or engineered sophistication.

Nat Neurosci. 2019 Jan;22(1):134-142. (PMID: 30559477)
Behav Brain Sci. 2017 Jan;40:e253. (PMID: 27881212)
Behav Brain Sci. 2009 Feb;32(1):69-84; discussion 85-120. (PMID: 19210833)
Nature. 2013 Nov 7;503(7474):78-84. (PMID: 24201281)
Nat Neurosci. 2013 Sep;16(9):1170-8. (PMID: 23955561)
Science. 1974 Sep 27;185(4157):1124-31. (PMID: 17835457)
Neuron. 2016 Dec 21;92(6):1398-1411. (PMID: 27916454)
Nat Neurosci. 2019 Dec;22(12):2066-2077. (PMID: 31659343)
Nat Neurosci. 2020 Sep;23(9):1138-1149. (PMID: 32778794)
Learn Mem. 2016 Jan 19;23(2):90-8. (PMID: 26787780)
Proc Natl Acad Sci U S A. 2020 Nov 24;117(47):29381-29389. (PMID: 33229518)
Nat Neurosci. 2011 Jun;14(6):783-90. (PMID: 21552276)
Science. 2018 May 4;360(6388):537-542. (PMID: 29567809)
Nat Neurosci. 2018 Jun;21(6):860-868. (PMID: 29760527)
Nat Rev Neurosci. 2021 Jan;22(1):55-67. (PMID: 33199854)
Science. 1996 Sep 6;273(5280):1399-402. (PMID: 8703077)
Nat Commun. 2019 Aug 21;10(1):3770. (PMID: 31434893)
Nat Neurosci. 2019 Nov;22(11):1761-1770. (PMID: 31659335)
J Neurophysiol. 2007 Sep;98(3):1125-39. (PMID: 17615138)
Nat Rev Neurosci. 2006 May;7(5):358-66. (PMID: 16760916)
Nature. 2001 Nov 29;414(6863):546-50. (PMID: 11734855)
Trends Neurosci. 2004 Dec;27(12):712-9. (PMID: 15541511)
PLoS Comput Biol. 2019 Apr 9;15(4):e1006972. (PMID: 30964861)
Nat Neurosci. 2008 Jan;11(1):95-102. (PMID: 18066060)
Nat Rev Neurosci. 2008 Apr;9(4):292-303. (PMID: 18319728)
Nature. 2021 Mar;591(7851):604-609. (PMID: 33473215)
Neuron. 2020 Feb 5;105(3):416-434. (PMID: 32027833)
Nat Neurosci. 2007 Sep;10(9):1214-21. (PMID: 17676057)
Nature. 2007 Jun 28;447(7148):1075-80. (PMID: 17546027)
Nat Neurosci. 2015 Oct;18(10):1509-17. (PMID: 26343249)
Nat Hum Behav. 2021 Jan;5(1):99-112. (PMID: 33168951)
Sci Adv. 2023 Mar 29;9(13):eadd0501. (PMID: 36989365)
Neuron. 2011 Jun 23;70(6):1054-69. (PMID: 21689594)
Nat Neurosci. 2007 May;10(5):647-56. (PMID: 17450137)
Cognition. 2009 Dec;113(3):262-280. (PMID: 18926527)
Elife. 2021 Dec 02;10:. (PMID: 34854377)
Neuron. 2013 Oct 16;80(2):519-30. (PMID: 24139048)

Date Created: 20241030 Date Completed: 20241030 Latest Revision: 20241101

20241101

PMC11524185

10.1126/sciadv.adl3931

39475619