Deep active inference


This work combines the free energy principle and the ensuing active inference dynamics with recent advances in variational inference in deep generative models, and evolution strategies to introduce the “deep active inference” agent. This agent minimises a variational free energy bound on the average surprise of its sensations, which is motivated by a homeostatic argument. It does so by optimising the parameters of a generative latent variable model of its sensory inputs, together with a variational density approximating the posterior distribution over the latent variables, given its observations, and by acting on its environment to actively sample input that is likely under this generative model. The internal dynamics of the agent are implemented using deep and recurrent neural networks, as used in machine learning, making the deep active inference agent a scalable and very flexible class of active inference agent. Using the mountain car problem, we show how goal-directed behaviour can be implemented by defining appropriate priors on the latent states in the agent’s model. Furthermore, we show that the deep active inference agent can learn a generative model of the environment, which can be sampled from to understand the agent’s beliefs about the environment and its interaction therewith.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8
Fig. 9
Fig. 10
Fig. 11


  1. 1.

    In more general formulations of active inference, the assumption that the mapping between hidden states and outcomes is constant can be relaxed (Friston et al. 2015).


  1. Adams RA, Stephan KE, Brown H, Frith CD, Friston KJ (2013) The computational anatomy of psychosis. Front Psychiatry 4:47

    Article  Google Scholar 

  2. Alais D, Burr D (2004) The ventriloquist effect results from near-optimal bimodal integration. Curr Biol 14(3):257–262

    CAS  Article  Google Scholar 

  3. Baez JC, Pollard BS (2015) Relative entropy in biological systems. arXiv:1512.02742

  4. Baltieri M, Buckley CL (2017) An active inference implementation of phototaxis. arXiv:1707.01806

  5. Berkes P, Orbán G, Lengyel M, Fiser J (2011) Spontaneous cortical activity reveals hallmarks of an optimal internal model of the environment. Science 331:83–87

    CAS  Article  Google Scholar 

  6. Brockman G, Cheung V, Pettersson L, Schneider J, Schulman J, Tang J, Zaremba W (2016) Openai gym. arXiv:1606.01540

  7. Brown H, Friston KJ (2012) Free-energy and illusions: the cornsweet effect. Front Psychol 3:43

    PubMed  PubMed Central  Google Scholar 

  8. Campbell JO (2016) Universal Darwinism as a process of Bayesian inference. arXiv:1606.07937

  9. Caticha A (2004) Relative entropy and inductive inference. In: AIP conference proceedings, 707

  10. Chung J, Kastner K, Dinh L, Goel K, Courville A, Bengio Y (2015) A recurrent latent variable model for sequential data. arXiv:1506.02216

  11. Conant R, Ashby W (1970) Every good regulator of a system must be a model of that system. Int J Syst Sci 1(2):89–97

    Article  Google Scholar 

  12. Crapse TB, Sommer MA (2008) Corollary discharge across the animal kingdom. Nat Rev Neurosci 9:587–600

    CAS  Article  Google Scholar 

  13. Dosovitskiy A, Koltun V (2017) Learning to act by predicting the future. ICLR

  14. Erhan D, Bengio Y, Courville A, Manzagol PA, Vincent P (2010) Why does unsupervised pre-training help deep learning? JMLR 11:625–660

    Google Scholar 

  15. Ernst M, Banks M (2002) Humans integrate visual and haptic information in a statistically optimal fashion. Nature 415:429–433

    CAS  Article  Google Scholar 

  16. Friston KJ (2005) A theory of cortical responses. Phil Trans R Soc B 360:815–836

    Article  Google Scholar 

  17. Friston KJ (2008) Hierarchical models in the brain. PLoS Comput Biol 4(11):e1000211

    Article  Google Scholar 

  18. Friston KJ (2010) The free-energy principle: a unified brain theory? Nat Rev Neurosci 11(2):127–138

    CAS  Article  Google Scholar 

  19. Friston KJ (2012) A free energy principle for biological systems. Entropy 14:2100–2121

    Article  Google Scholar 

  20. Friston KJ (2013) Life as we know it. J R Soc Interface 10:20130475

    Article  Google Scholar 

  21. Friston KJ, Kiebel SJ (2009) Predictive coding under the free-energy principle. Philos Trans R Soc B 364:1211–1221

    Article  Google Scholar 

  22. Friston KJ, Kilner J, Harrison L (2006) A free energy principle for the brain. J Physiol Paris 100:70–87

    Article  Google Scholar 

  23. Friston KJ, Daunizeau J, Kilner J, Kiebel SJ (2010) Action and behavior: a free-energy formulation. Biol Cybern 192(3):227–260

    Article  Google Scholar 

  24. Friston KJ, Mattout J, Kilner J (2011) Action understanding and active inference. Biol Cybern 104:137–160

    Article  Google Scholar 

  25. Friston KJ, Rigoli F, Ognibene D, Mathys C, Fitzgerald T, Pezzulo G (2015) Active inference and epistemic value. Cogn Neurosci 6(4):187–214

    Article  Google Scholar 

  26. Friston KJ, Frith CD, Pezzulo G, Hobson JA, Ondobaka S (2017a) Active inference, curiosity and insight. Neural Comput 29:1–51

    Article  Google Scholar 

  27. Friston KJ, Rosch R, Parr T, Price C, Bowman H (2017b) Deep temporal models and active inference. Neurosci Biobehav Rev 77:388–402

    Article  Google Scholar 

  28. Goodfellow I, Pouget-Abadie J, Mirza M, Xu B, Warde-Farley D, Ozair S, Courville A, Bengio Y (2014) Generative adversarial networks. arXiv:1406.2661

  29. Goodfellow I, Bengio Y, Courville A (2016) Deep Learning. MIT Press, Cambridge.

  30. Graves A, Wayne G, Reynolds M, Harley T, Danihelka I, Grabska-Barwinska A, Gómez Caolmenarejo S, Grefenstette E, Ramalho T, Agapiou J, Puigdomenèch Badia A, Hermann KM, Zwols Y, Ostrovski G, Cain A, King H, Summerfield C, Blunsum P, Kavukcuoglu K, Hassabis D (2016) Hybrid computing using a neural network with dynamic external memory. Nature 538:471–476

    Article  Google Scholar 

  31. Ha D, Schmidhuber J (2018) World models. arXiv:1803.10122

  32. Haefner R, Berkes P, Fiser J (2016) Perceptual decision-making as probabilistic inference by neural sampling. Neuron 90(3):649–660

    CAS  Article  Google Scholar 

  33. Hansen N (2016) The CMA evolution strategy: a tutorial. arXiv:1604.00772

  34. Harper M (2009) The replicator equation as an inference dynamic. arXiv:0911.1763

  35. Hinton GE, Salakhutdinov RR (2006) Reducing the dimensionality of data with neural networks. Science 313:504–507

    CAS  Article  Google Scholar 

  36. Hornik K, Stinchcombe M, White H (1989) Multilayer feedforward networks are universal approximators. Neural Netw 2:359–366

    Article  Google Scholar 

  37. Huszár F (2017) Variational inference using implicit distributions. arXiv:1702.08235

  38. Karpathy A, Johnson J, Fei-Fei L (2015) Visualizing and understanding recurrent networks. arXiv:1506.02078

  39. Karras T, Aila T, Laine S, Lehtinen J (2018) Progressive growing of gans for improved quality, stability, and variation. ICLR

  40. Kempka M, Wydmuch M, Runc G, Toczek J, Jaśkowski W (2016) Vizdoom: a doom-based AI research platform for visual reinforcement learning. arXiv:1605.02097

  41. Kingma DP, Ba J (2014) Adam: a method for stochastic optimization. arXiv:1412.6980

  42. Kingma DP, Welling M (2014) Auto-encoding variational bayes. ICLR

  43. Kingma DP, Salimans T, Jozefowicz R, Chen X, Sutskever I, Welling M (2016) Improving variational inference with inverse autoregressive flow. arXiv:1606.04934

  44. Knill D, Pouget A (2004) The bayesian brain: the role of uncertainty in neural coding and computation. Trends Neurosci 27(12):712–719

    CAS  Article  Google Scholar 

  45. Le QV, Jaitly N, Hinton GE (2015) A simple way to initialize recurrent networks of rectified linear units. arXiv:1504.00941

  46. LeCun Y, Bengio Y, Hinton GE (2015) Deep learning. Nature 521:436–444

    CAS  Article  Google Scholar 

  47. Ledig C, Theis L, Huszár F, Caballero J, Cunningham A, Acosta A, Aitken A, Tejani A, Totz J, Wang Z, Shi W (2016) Photo-realistic single image super-resolution using a generative adversarial network. arXiv:1609.04802

  48. Liu MY, Breuel T, Kautz J (2017) Unsupervised image-to-image translation networks. Neural Information Processing Systems (NIPS). arXiv:1703.00848

  49. Maaløe L, Sønderby CK, Sønderby SK, Winther O (2016) Auxiliary deep generative models. arXiv:1602.05473

  50. Maheswaranathan N, Metz L, Tucker G, Sohl-Dickenstein J (2018) Guided evolutionary strategies: escaping the curse of dimensionality in random search. arXiv:1806.10230

  51. Mescheder L, Nowozin S, Geiger A (2017) Adversarial variational Bayes: unifying variational autoencoders and generative adversarial networks. arXiv:1701.04722

  52. Mnih V, Kavukcuoglu K, Silver D, Rusu AA, Veness J, Bellemare MG, Graves A, Riedmiller M, Fidjeland AK, Ostrovski G, Petersen S, Beattie C, Sadik A, Antonoglou I, King H, Kumaran D, Wierstra D, Legg S, Hassabis D (2015) Human-level control through deep reinforcement learning. Nature 518:529–533

    CAS  Article  Google Scholar 

  53. Moore A (1991) Variable resolution dynamic programming: efficiently learning action maps in multivariate real-valued state-spaces. In: Proceedings of the eight international conference on machine learning. Morgan Kaufmann

  54. Moreno-Bote R, Knill D, Pouget A (2011) Bayesian sampling in visual perception. Proc Natl Acad Sci USA 108(30):12491–12496

    CAS  Article  Google Scholar 

  55. Pathak D, Pulkit A, Efros AA, Darrell T (2017) Curiosity-driven exploration by self-supervised prediction. arXiv:1705.05363

  56. Platt JC, Barr AH (1988) Constrained differential optimization. In: Neural information processing systems. American Institute of Physics, New York, pp 612–621

  57. Radford A, Narasimhan K, Salimans T, Sutskever I (2018) Improving language understanding by generative pre-training. Technical report, OpenAI

  58. Ramstead MJD, Badcock PB, Friston KJ (2017) Answering schrödinger’s question: a free-energy formulation. Phys Life Rev 24:1–16

    Article  Google Scholar 

  59. Rezende DJ, Mohamed S (2015) Variational inference with normalizing flows. JMLR 37

  60. Rezende DJ, Mohamed S, Wierstra D (2014) Stochastic backpropagation and approximate inference in deep generative models. ICML

  61. Rezende DJ, Ali Eslami SM, Mohamed S, Battaglia P, Jaderberg M, Heess N (2016) Unsupervised learning of 3d structure from images. arXiv:1607.00662

  62. Salimans T, Ho J, Chen X, Sutskever I (2017) Evolution strategies as a scalable alternative to reinforcement learning. arXiv:1703.03864

  63. Schwartenbeck P, Fitzgerald T, Mathys C, Dolan R, Kronbichler M, Friston KJ (2015) Evidence for surprise minimization over value maximization in choice behavior. Sci Rep 5:16575

    Article  Google Scholar 

  64. Siegelmann HT (1995) Computation beyond the turing limit. Science 268:545–548

    CAS  Article  Google Scholar 

  65. Theano Development Team (2016) Theano: a Python framework for fast computation of mathematical expressions. arXiv:1605.02688

  66. Todorov E, Erez T, Tassa Y (2012) Mujoco: A physics engine for model-based control. In: Proceedings of the IEEE/RSJ international conference on intelligent robots and systems (IROS)

  67. Tomczak JM, Welling M (2016) Improving variational auto-encoders using householder flow. arXiv:1611.09630

  68. Tran D, Ranganath R, Blei D (2017) Hierarchical implicit models and likelihood-free variational inference. arXiv:1702.08896

  69. Watson RA, Szathmáry E (2016) How can evolution learn? Trends Ecol Evol 31(2):147–157

    Article  Google Scholar 

  70. Wong KF, Wang XJ (2006) A recurrent network mechanism of time integration in perceptual decisions. J Neurosci 26(4):1314–1328

    CAS  Article  Google Scholar 

  71. Zhu JY, Park T, Isola P, Efros AA (2017) Unpaired image-to-image translation using cycle-consistent adversarial networks. arXiv:1703.10593

Download references


The author would like to thank Karl Friston, Thorben Kröger, Manuel Baltieri and Annina Luck for insightful comments on earlier versions of this manuscript and the participants and organisers of the Computational Psychiatry Course 2016 for stimulating lectures and discussions. Furthermore, he would like to thank the three anonymous reviewers for very constructive and tremendously helpful comments, which significantly improved the quality of the paper.

Author information



Corresponding author

Correspondence to Kai Ueltzhöffer.

Additional information

Communicated by Benjamin Lindner.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary material 1 (pdf 542 KB)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Ueltzhöffer, K. Deep active inference. Biol Cybern 112, 547–573 (2018).

Download citation


  • Variational inference
  • Deep learning
  • Generative models
  • Cognition
  • Action
  • Perception