1. 2. 3. Estimation of parameters in iFEP
The ReCU model has several parameters: σ2w, α, β, Po and ϵ. In the estimation, we set ϵ to 1, which was the optimal value for estimation in the
artificial data (Supplementary Fig. 3). We assumed the unit intensity
of reward, that is, ln Po /(1 − Po ) = 1 , because it is impossible to estimate
both Po and β caused by multiplying β and ln Po /(1 − Po ) in the expected
net utility (equations (30) and (42)). This treatment is suitable for relative comparison between the curiosity meta-parameter and the reward.
In addition, we addressed βct as a latent variable as ĉt = βct because of
the multiplication of β in the expected net utility (equations (30)
and (42)). Thus, the estimation of ct can be obtained by dividing the
estimated ĉt by the estimated β. Therefore, the hyperparameters to be
estimated were σ2w, α and β.
To estimate these parameters θ = {σ2w , α, β} , we extended the
observer-SSM to a self-organizing SSM44 in which θ was addressed as
constant latent variables:
P (zt , θ|x1∶t ) ∝ P (xt |zt ) ∫P (zt |zt−1 , θ) P (zt−1 , θ|x1∶t−1 ) dzt−1 ,
(78)
where P (θ) = Uni (σ2 |aσ , bσ ) Uni(α|aα , bα )𝒩𝒩𝒩β|mβ , vβ ). To sequentially calculate the posterior P (zt , θ|x1∶t ) using the particle filter, we used
100,000 particles and augmented the state vector of all particles by
adding the parameter θ, which was not updated from randomly sampled initial values.
The hyperparameter values used in this estimation were μ0 = 0,
σ2μ = 0.012 , ag = 10, bg = 0.001, au = −15, bu = 15, aσ = 0.2, bσ = 0.7, aα = 0.04,
bα = 0.06, aβ = 0 and bβ = 50, which were heuristically given as parameters
correctly estimated using the artificial data (Supplementary Fig. 2).
4. 5. 6. 7. 8. 9. 10. 11. 12. 13. Statistical testing with Monte Carlo simulations
Supplementary Fig. 5 shows statistical testing of the negative curiosity
estimated in Fig. 5. A null hypothesis is that an agent has no curiosity
(that is, ct = 0) decides on a choice only depending on its recognition of
the reward probability. Under the null hypothesis, model simulations
were repeated 1,000 times under the same experimental conditions as
in Fig. 5 and the curiosity was estimated for each using iFEP. We adopted
the temporal average of the estimated curiosity as a test statistic and
plotted the null distribution of the test statistic. Compared with the
estimated curiosity of the rat behavior, we computed the P value for a
one-sided left-tailed test.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
Source data for Figs. 2, 3, 5 and 6 are available with this paper. Source
data for Supplementary Figures are available in Supplementary Data.
Nature Computational Science | Volume 3 | May 2023 | 418–432
14. 15. 16. 17. 18. 19. 20. 21. Helmholtz, H. Handbuch der Physiologischen Optik (Andesite
Press, 1867).
Yuille, A. & Kersten, D. Vision as Bayesian inference: analysis by
synthesis? Trends Cogn. Sci. 10, 301–308 (2006).
Millett, J. D. & Simon, H. A. Administrative behavior: a study of
decision-making processes in administrative organization.
Polit. Sci. Q. 62, 621 (1947).
Dubey, R. & Griffiths, T. L. Understanding exploration in humans
and machines by formalizing the function of curiosity. Curr. Opin.
Behav. Sci. 35, 118–124 (2020).
Kidd, C. & Hayden, B. Y. The psychology and neuroscience of
curiosity. Neuron 88, 449–460 (2015).
Klein, U. & Nowak, A. J. Characteristics of patients with autistic
disorder (AD) presenting for dental treatment: a survey and chart
review. Spec. Care Dentist. 19, 200–207 (1999).
Lockner, D. W., Crowe, T. K. & Skipper, B. J. Dietary intake and
parents’ perception of mealtime behaviors in preschoolage children with autism spectrum disorder and in typically
developing children. J. Am. Diet. Assoc. 108, 1360–1363 (2008).
Schreck, K. A. & Williams, K. Food preferences and factors
influencing food selectivity for children with autism spectrum
disorders. Res. Dev. Disabil. 27, 353–363 (2006).
Esposito, M. et al. Sensory processing, gastrointestinal symptoms
and parental feeding practices in the explanation of food
selectivity: clustering children with and without autism.
Int. J. Autism Relat. Disabil. 2, 1–12 (2019).
Hobson, R. P. Autism and the development of mind. Essays Dev.
Psychol. (Routledge, 1993).
Burke, R. Personalized recommendation of PoIs to people with
autism. Commun. ACM 65, 100 (2022).
Ghanizadeh, A. Educating and counseling of parents of children
with attention-deficit hyperactivity disorder. Patient Educ. Couns.
68, 23–28 (2007).
Sedgwick, J. A., Merwood, A. & Asherson, P. The positive
aspects of attention deficit hyperactivity disorder: a qualitative
investigation of successful adults with ADHD. ADHD Atten. Deficit
Hyperact. Disord. 11, 241–253 (2019).
Redshaw, R. & McCormack, L. ‘Being ADHD’: a qualitative study.
Adv. Neurodev. Disord. 6, 20–28 (2022).
Sutton, R. S. & Barto, A. G. Reinforcement Learning: an Introduction
(MIT Press, 1998).
Friston, K. A theory of cortical responses. Philos. Trans. R. Soc. B
360, 815–836 (2005).
Friston, K., Kilner, J. & Harrison, L. A free energy principle for the
brain. J. Physiol. Paris 100, 70–87 (2006).
Friston, K. The free-energy principle: a unified brain theory?
Nat. Rev. Neurosci. 11, 127–138 (2010).
Lindley, D. V. On a measure of the information provided by an
experiment. Ann. Math. Stat. 27, 986–1005 (1956).
MacKay, D. J. C. Information-based objective functions for active
data selection. Neural Comput. 4, 590–604 (1992).
Berger, J. O. Statistical Decision Theory and Bayesian Analysis,
Springer Series in Statistics (Springer, 2011).
431
Article
22. Friston, K. et al. Active inference and epistemic value. Cogn.
Neurosci. 6, 187–214 (2015).
23. Friston, K., FitzGerald, T., Rigoli, F., Schwartenbeck, P. & Pezzulo, G.
Active inference: a process theory. Neural Comput. 29, 1–49 (2017).
24. Attias, H. Planning by probabilistic inference in Proc. 9th Int. Work.
Artif. Intell. Stat. 4, 9–16 (2003).
25. Botvinick, M. & Toussaint, M. Planning as inference. Trends Cogn.
Sci. 16, 485–488 (2012).
26. Kaplan, R. & Friston, K. J. Planning and navigation as active
inference. Biol. Cybern. 112, 323–343 (2018).
27. Matsumoto, T. & Tani, J. Goal-directed planning for habituated
agents by active inference using a variational recurrent neural
network. Entropy 22, (2020).
28. Schwartenbeck, P. et al. Computational mechanisms of curiosity
and goal-directed exploration. eLife 8, 1–45 (2019).
29. Millidge, B., Tschantz, A. & Buckley, C. L. Whence the expected
free energy? Neural Comput. 33, 447–482 (2021).
30. Houthooft, R. et al. VIME: variational information maximizing
exploration. Adv. Neural Inf. Process. Syst. 0, 1117–1125 (2016).
31. Smith, R. et al. Greater decision uncertainty characterizes a
transdiagnostic patient sample during approach-avoidance
conflict: a computational modelling approach. J. Psychiatry
Neurosci. 46, E74–E87 (2021).
32. Smith, R. et al. Long-term stability of computational parameters
during approach-avoidance conflict in a transdiagnostic
psychiatric patient sample. Sci Rep. 11, 1–13 (2021).
33. Schwartenbeck, P. & Friston, K. Computational phenotyping in
psychiatry: a worked example. eNeuro 3, 1–18 (2016).
34. Daunizeau, J. et al. Observing the observer (I): meta-Bayesian models
of learning and decision-making. PLoS ONE 5, e15554 (2010).
35. Patzelt, E. H., Hartley, C. A. & Gershman, S. J. Computational
phenotyping: using models to understand individual differences
in personality, development, and mental illness. Personal.
Neurosci. 1, e18 (2018).
36. Ito, M. & Doya, K. Validation of decision-making models and
analysis of decision variables in the rat basal ganglia. J. Neurosci.
29, 9861–9874 (2009).
37. Samejima, K., Doya, K., Ueda, Y. & Kimura, M. Estimating internal
variables and parameters of a learning agent by a particle filter.
Adv. Neural Inf. Process. Syst. 16 (2003).
38. Samejima, K., Ueda, Y., Doya, K. & Kimura, M. Neuroscience:
representation of action-specific reward values in the striatum.
Science (80-.) 310, 1337–1340 (2005).
39. Ortega, P. A. & Braun, D. A. Thermodynamics as a theory of
decision-making with information-processing costs. Proc. R. Soc.
London. A 469, 20120683 (2013).
40. Gottwald, S. & Braun, D. A. The two kinds of free energy and the
Bayesian revolution. PLoS Comput. Biol. 16, (2020).
41. Parr, T. & Friston, K. J. Generalised free energy and active
inference. Biol. Cybern. 113, 495–513 (2019).
42. Kitagawa, G. Monte Carlo filter and smoother for non-Gaussian
nonlinear state space models. J. Comput. Graph. Stat. 5, 1–25 (1996).
43. Bishop, C. M. Pattern Recognition and Machine Learning
(Springer, 2006).
44. Kitagawa, G. A self-organizing state-space model. J. Am. Stat. 93,
1203–1215 (1998).
45. Konaka, Y. & Naoki, H. Codes for Konaka and Honda 2023. Zenodo
https://doi.org/10.5281/zenodo.7722905 (2023)
Nature Computational Science | Volume 3 | May 2023 | 418–432
https://doi.org/10.1038/s43588-023-00439-w
Acknowledgements
We are grateful to K. Doya and M. Ito for providing rat behavioral
data. We thank the organizers of the tutorial on the free energy
principle in 2019, which inspired this research, and I. Higashino
and M. Fujiwara-Yada for carefully checking all the equations in
the manuscript. This study was supported in part by a Grant-in-Aid
for Transformative Research Areas (B) (no. 21H05170), AMED
(grant no. JP21wm0425010), Moonshot R&D–MILLENNIA program
(grant no. JPMJMS2024-9) by JST, the Cooperative Study
Program of Exploratory Research Center on Life and Living Systems
(ExCELLS) (program no. 21-102) and the grant of Joint Research
by the National Institutes of Natural Sciences (NINS program no.
01112102).
Author contributions
H.N. conceived of the project. Y. K. and H.N. developed the method,
and Y.K. implemented the model simulation. Y.K. and H.N. wrote
the manuscript.
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information The online version
contains supplementary material available at
https://doi.org/10.1038/s43588-023-00439-w.
Correspondence and requests for materials should be addressed to
Honda Naoki.
Peer review information Nature Computational Science thanks
Junichiro Yoshimoto and Karl Friston for their contribution to the
peer review of this work. Primary Handling Editors: Ananya Rastogi
and Jie Pan, in collaboration with the Nature Computational Science
team.
Reprints and permissions information is available at
www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with
regard to jurisdictional claims in published maps and
institutional affiliations.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format,
as long as you give appropriate credit to the original author(s) and the
source, provide a link to the Creative Commons license, and indicate
if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons license, unless
indicated otherwise in a credit line to the material. If material is not
included in the article’s Creative Commons license and your intended
use is not permitted by statutory regulation or exceeds the permitted
use, you will need to obtain permission directly from the copyright
holder. To view a copy of this license, visit http://creativecommons.
org/licenses/by/4.0/.
© The Author(s) 2023
432
...