CoChat

Sophisticated Inference

Open access
Friston, Karl, Da Costa, Lancelot, Hafner, Danijar, Hesp, Casper, Parr, Thomas · 2021 · Neural Computation
Cited by 117

The DOI matches its Crossref record; 33 of the 33 references it cites were found in a registry or an index.

Abstract

Active inference offers a first principle account of sentient behavior, from which special and important cases—for example, reinforcement learning, active learning, Bayes optimal inference, Bayes optimal design—can be derived. Active inference finesses the exploitation-exploration dilemma in relation to prior preferences by placing information gain on the same footing as reward or value. In brief, active inference replaces value functions with functionals of (Bayesian) beliefs, in the form of an expected (variational) free energy. In this letter, we consider a sophisticated kind of active inference using a recursive form of expected free energy. Sophistication describes the degree to which an agent has beliefs about beliefs. We consider agents with beliefs about the counterfactual consequences of action for states of affairs and beliefs about those latent states. In other words, we move from simply considering beliefs about “what would happen if I did that” to “what I would believe about what would happen if I did that.” The recursive form of the free energy functional effectively implements a deep tree search over actions and outcomes in the future. Crucially, this search is over sequences of belief states as opposed to states per se. We illustrate the competence of this scheme using numerical simulations of deep decision problems.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1162/neco_a_01351

Cites

33 of 33 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 33 references

Research works printed but not linked: 75

  1. 10.1162/089976698300017746 · DOI · Paper page
  2. Attias H., 2003, Proceedings of the 9th Int. Workshop on Artificial Intelligence and Statistics
  3. 10.3390/e17042432 · DOI · Paper page
  4. Barlow H., 1961, Sensory communication, 217
  5. 10.1068/p030123 · DOI · Paper page
  6. Beal M. J., 2003, Variational algorithms for approximate Bayesian inference
  7. 10.1162/089976601753195969 · DOI · Paper page
  8. 10.1016/j.tics.2012.08.006 · DOI · Paper page
  9. 10.1162/0033553041502225 · DOI · Paper page
  10. Catal O., 2019, Bayesian policy selection using active inference
  11. Çatal O., 2020, Proceedings of the 2020 IEEE International Conference on Acoustics, Speech and Signal Processing
  12. Çatal O., 2020, Deep active inference for autonomous robot navigation
  13. 10.1063/1.4905960 · DOI
  14. 10.1111/1468-0262.00239 · DOI · Paper page
  15. 10.1103/PhysRevLett.99.100602 · DOI · Paper page
  16. Da Costa L., 2020, Active inference on discrete state-spaces: A synthesis
  17. Da Costa L., 2020, Natural selection finds natural gradient
  18. Da Costa L., 2020, The relationship between dynamic programming and active inference: The discrete, finite-horizon case
  19. 10.1016/j.neuron.2011.02.027 · DOI · Paper page
  20. 10.1162/neco.1993.5.4.613 · DOI · Paper page
  21. 10.1162/neco.1995.7.5.889 · DOI · Paper page
  22. 10.1371/journal.pone.0087619 · DOI · Paper page
  23. 10.1016/S0893-6080(99)00046-5 · DOI · Paper page
  24. Duff M. O., 2002, Optimal learning: Computational procedures for Bayes-adaptive Markov decision processes
  25. 10.1214/aoap/1026915623 · DOI
  26. 10.1162/neco_a_00999 · DOI · Paper page
  27. 10.1016/j.neubiorev.2017.04.009 · DOI · Paper page
  28. 10.1016/j.cobeha.2017.05.025 · DOI · Paper page
  29. 10.1023/A:1007425814087 · DOI · Paper page
  30. Ghavamzadeh M., 2016, Bayesian reinforcement learning: A survey
  31. Hafez-Kolahi H., 2019, Information bottleneck and its applications in deep learning
  32. 10.1109/TSSC.1966.300074 · DOI · Paper page
  33. 10.1162/0899766041336477 · DOI · Paper page
  34. Jaynes E. T., 1957, Physical Review Series II, 106, 620
  35. 10.1007/s00422-018-0753-2 · DOI · Paper page
  36. 10.2307/2226451 · DOI · Paper page
  37. 10.1073/pnas.1609094113 · DOI · Paper page
  38. 10.3390/e20080574 · DOI · Paper page
  39. Klyubin A. S., 2005, Proceedings of the IEEE Congress on Evolutionary Computation, 1:128
  40. 10.1371/journal.pcbi.1005753 · DOI
  41. 10.1146/annurev.ne.13.030190.001353 · DOI · Paper page
  42. MacKay D. J. C., 2003, Information theory, inference and learning algorithms
  43. Millidge B., 2019, Deep active inference as variational policy gradients
  44. 10.1038/s41562-017-0180-8 · DOI · Paper page
  45. 10.1152/jn.1987.57.1.162 · DOI · Paper page
  46. 10.1613/jair.3062 · DOI · Paper page
  47. Osband I., 2019, Journal of Machine Learning Research, 20, 1
  48. Parr T., 2020, Philosophical Transactions of the Royal Society A, 378
  49. 10.1038/s41598-017-15249-0 · DOI · Paper page
  50. 10.1016/j.copsyc.2018.10.006 · DOI · Paper page
  51. Rikhye R. V., 2019, Memorize-generalize: An online algorithm for learning higher-order sequential structure with cloned hidden Markov models
  52. Ross S., 2008, Advances in neural information processing systems, 20
  53. 10.1371/journal.pcbi.1005768 · DOI · Paper page
  54. Ryan R., 1985, Intrinsic motivation and self-determination in human behavior
  55. 10.1109/IJCNN.1991.170605 · DOI · Paper page
  56. 10.1080/09540090600768658 · DOI · Paper page
  57. Schrittwieser J., 2019, Mastering Atari, go, chess and shogi by planning with a learned model
  58. 10.3389/fpsyg.2013.00710 · DOI · Paper page
  59. 10.7554/eLife.41703 · DOI · Paper page
  60. Sengupta, B. & Friston, K. (2018). How robust are deep neural networks? arXiv:1804.11313. · arXiv
  61. 10.1073/pnas.1505483112 · DOI · Paper page
  62. 10.1103/PhysRevLett.109.120604 · DOI · Paper page
  63. 10.1109/IJCNN.2016.7727309 · DOI · Paper page
  64. Sutton R. S., 1998, Reinforcement learning: An introduction
  65. Tishby N., 1999, Proceedings of the 37th Annual Allerton Conference on Communication, Control, and Computing, 368
  66. Tishby N., 2010, Perception-reason-action cycle: Models, algorithms and systems
  67. Tishby N., 2015, Deep learning and the information bottleneck principle
  68. Todorov E., 2008, Proceedings of the IEEE Conference on Decision and Control
  69. Toussaint M., 2006, Proceedings of the 23rd Int. Conf. on Machine Learning
  70. Tribus M., 1961, Thermodynamics and thermostatics: An introduction to energy, information and states of matter, with engineering applications
  71. Tschantz A., 2019, Scaling active inference
  72. van den Broek J. L., 2010, Uncertainty in Artificial Intelligence, 6, 1
  73. Von Neumann J., 1944, Theory of games and economic behavior
  74. Winn J., 2005, Journal of Machine Learning Research, 6, 661
  75. 10.1109/TIT.2005.850085 · DOI · Paper page

Cited by

Metadata

Authors
Friston, K., Da Costa, L., Hafner, D., Hesp, C., & Parr, T.
Venue
Neural Computation
Published
2021-03
Volume
33
Issue
3
Pages
713-763
ISSN
  • 0899-7667 (ISSN-L)
  • 1530-888X
OpenAlex
W3033196763
Cited by
117
References
33
FWCI
10.98
Type
journal article
Language
en
Open access
green

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.