pith. sign in

arxiv: 2606.20658 · v1 · pith:TJC6W2DFnew · submitted 2026-06-09 · 💻 cs.AI · cs.LG

Expected Free Energy-based Planning as Variational Inference

Wouter W. L. Nuijten , Thijs van de Laar , Bert de Vries This is my paper

Pith reviewed 2026-06-27 13:37 UTC · model grok-4.3

classification 💻 cs.AI cs.LG
keywords expected free energyvariational free energyactive inferenceplanning under uncertaintyepistemic priorsfree energy principlevariational inference
0
0 comments X

The pith

Minimizing variational free energy with epistemic priors decomposes into expected free energy plan costs plus a complexity term.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that expected free energy planning can be recast as standard variational free energy minimization once the generative model includes appropriately chosen epistemic priors. This treats planning as one more instance of the same inferential process that handles perception and learning under the free energy principle. A reader would care because it removes the need for specialized optimization routines and lets existing variational inference tools apply directly to planning problems. Experiments on T-maze, stochastic maze, and MiniGrid tasks confirm that the resulting policies exhibit information-seeking behavior and scale beyond tabular methods.

Core claim

By augmenting the generative model with epistemic priors, the variational free energy functional decomposes into the expected free energy (which encodes both instrumental goal costs and epistemic information value) plus an explicit complexity term. This equivalence demonstrates that EFE-based planning is variational inference on that augmented model, thereby aligning planning with the same free-energy minimization that governs other active-inference operations.

What carries the argument

Epistemic priors on the generative model that produce the exact decomposition of variational free energy into expected free energy plus complexity.

Load-bearing premise

It is possible to choose epistemic priors for the generative model such that variational free energy minimization produces exactly the expected free energy objective plus an additional complexity term.

What would settle it

If the policies obtained by minimizing the proposed variational free energy on the T-maze do not match the policies obtained by direct expected free energy optimization, the claimed decomposition does not hold.

Figures

Figures reproduced from arXiv: 2606.20658 by Bert de Vries, Thijs van de Laar, Wouter W. L. Nuijten.

Figure 1
Figure 1. Figure 1: T-maze environment. The agent starts at the center junction and must reach the rewarding arm. [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Representative trajectories with planned actions (reward in left arm). The blue circle indicates [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Variational Free Energy during optimization for each objective. T-maze (left), averaged over 20 [PITH_FULL_IMAGE:figures/full_fig_p020_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: T-maze: convergence curves for the first planning step with unknown parameters (left) and known [PITH_FULL_IMAGE:figures/full_fig_p021_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: T-maze: success rate as a function of optimization steps (left) and probability of the cue-visiting [PITH_FULL_IMAGE:figures/full_fig_p021_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Reactivity Maze: success rate as a function of optimization steps (left) and mean reward across [PITH_FULL_IMAGE:figures/full_fig_p022_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: T-maze: success rate as a function of learning rate (left) and mean reward across 5 random seeds [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Reactivity Maze: success rate as a function of learning rate. [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: MiniGrid DoorKey: convergence trajectories across knowledge scenarios (columns) and objectives [PITH_FULL_IMAGE:figures/full_fig_p025_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: MiniGrid DoorKey: convergence trajectories at planning horizons [PITH_FULL_IMAGE:figures/full_fig_p026_10.png] view at source ↗
read the original abstract

Planning under uncertainty requires agents to balance goal achievement with information gathering. Active inference addresses this through the Expected Free Energy (EFE), a cost function that unifies instrumental and epistemic objectives. However, existing EFE-based methods typically employ specialized optimization procedures that are difficult to extend or analyze. In this paper, we show that EFE-based planning can be formulated as Variational Free Energy minimization on a generative model augmented with epistemic priors. Our main result demonstrates that minimizing a Variational Free Energy functional with appropriately chosen priors yields a decomposition into expected plan costs (the EFE) plus a complexity term. This formulation reinforces theoretical consistency with the Free Energy Principle by casting planning as the same inferential process that governs perception and learning. We validate our approach on three environments of increasing complexity: a deterministic T-maze, a stochastic Reactivity Maze, and a partially observable MiniGrid DoorKey-8x8 environment. The experiments demonstrate that the epistemic priors induce information-seeking behavior, that the variational formulation yields policy-based inference outperforming plan-based methods under stochastic transitions, and that temporal factorization enables scalability to environments where existing tabular active inference methods cannot operate.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper claims that EFE-based planning can be recast as standard variational inference by augmenting a generative model with epistemic priors; minimizing the resulting variational free energy then decomposes exactly into the expected free energy (instrumental plus epistemic value) plus a complexity term. This is presented as reinforcing consistency with the Free Energy Principle. The claim is supported by a derivation and by experiments on a deterministic T-maze, a stochastic Reactivity Maze, and a partially observable MiniGrid DoorKey-8x8 task, where the method induces information-seeking behavior and scales to settings where tabular active-inference planners fail.

Significance. If the epistemic-prior construction is non-circular and holds for arbitrary transition and observation models, the result would allow active-inference planning to reuse existing variational-inference toolchains rather than requiring specialized EFE optimizers, thereby tightening the link between planning and the broader FEP framework. The reported scalability gains on the MiniGrid environment constitute a concrete empirical contribution.

major comments (2)
  1. [§3.2, Eq. (11)–(13)] §3.2, Eq. (11)–(13): the epistemic prior p(o, s | π) is defined so that the cross terms between the variational distribution and the generative model cancel, recovering the standard EFE decomposition. It is not shown that this prior can be specified from the environment dynamics alone without reference to the target EFE functional; if the prior is chosen to enforce the desired cancellation, the claimed equivalence to ordinary variational inference is circular.
  2. [§4.1] §4.1, the three-environment validation: while the T-maze and Reactivity Maze results demonstrate information-seeking, the MiniGrid experiment reports only aggregate success rates without an ablation that isolates the contribution of the epistemic prior versus the complexity term. Consequently it remains unclear whether the reported outperformance under stochastic transitions is due to the variational formulation or to the specific prior construction.
minor comments (2)
  1. [§2.3] Notation for the policy-conditioned generative model is introduced in §2.3 but reused without re-statement in §3; a short reminder of the conditioning would improve readability.
  2. [Figure 3] Figure 3 caption states “temporal factorization enables scalability” but does not quantify the computational saving relative to the non-factorized baseline; adding wall-clock or iteration counts would strengthen the claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which have helped us improve the clarity and rigor of our manuscript. We address each of the major comments point by point below, indicating the revisions we plan to make.

read point-by-point responses

  1. Referee: [§3.2, Eq. (11)–(13)] §3.2, Eq. (11)–(13): the epistemic prior p(o, s | π) is defined so that the cross terms between the variational distribution and the generative model cancel, recovering the standard EFE decomposition. It is not shown that this prior can be specified from the environment dynamics alone without reference to the target EFE functional; if the prior is chosen to enforce the desired cancellation, the claimed equivalence to ordinary variational inference is circular.

    Authors: We agree that the construction of the epistemic prior requires careful justification to avoid any appearance of circularity. In our derivation, the epistemic prior is defined to encode the expected information gain under the policy, which is a standard component of the EFE in active inference literature. This prior is specified using the transition and observation models of the environment, augmented with the epistemic objective. The equivalence to variational inference then follows directly, allowing the use of standard VI toolchains. To address the referee's concern, we will revise the manuscript in §3.2 to include an explicit construction of the prior solely from the generative model and the information-theoretic definition of epistemic value, without presupposing the full EFE. This will demonstrate that the approach holds for arbitrary models and is not circular. revision: yes

  2. Referee: [§4.1] §4.1, the three-environment validation: while the T-maze and Reactivity Maze results demonstrate information-seeking, the MiniGrid experiment reports only aggregate success rates without an ablation that isolates the contribution of the epistemic prior versus the complexity term. Consequently it remains unclear whether the reported outperformance under stochastic transitions is due to the variational formulation or to the specific prior construction.

    Authors: We acknowledge that the MiniGrid results would benefit from an ablation study to isolate the effects. The reported outperformance is attributed to the variational formulation enabling policy-based inference, which scales better than plan-based methods in stochastic settings. However, to clarify the role of the epistemic prior, we will add an ablation in the revised version comparing performance with and without the epistemic prior (i.e., setting the prior to uniform). This will include additional analysis of information-seeking behavior in the MiniGrid environment. revision: yes

Circularity Check

1 steps flagged

Priors selected to force VFE minimization to recover EFE + complexity term make the decomposition hold by construction

specific steps
  1. self definitional [Abstract]
    "Our main result demonstrates that minimizing a Variational Free Energy functional with appropriately chosen priors yields a decomposition into expected plan costs (the EFE) plus a complexity term."

    The priors are stipulated as 'appropriately chosen' exactly so that VFE minimization produces the EFE decomposition. This makes the claimed equivalence tautological: the functional form of the priors is reverse-engineered to cancel cross terms and recover the standard EFE expression, rather than being independently motivated and then shown to yield the result.

full rationale

The paper's main result states that augmenting the generative model with 'appropriately chosen' epistemic priors allows VFE minimization to decompose into the EFE plus a complexity term. This is self-definitional because the priors are defined precisely so that all cross terms cancel to yield the standard EFE (instrumental + epistemic value) without residual dependence on the variational distribution. The equivalence is therefore enforced by the choice of priors rather than derived from independent assumptions on arbitrary transition/observation models. No external verification or parameter-free justification is provided for the priors' existence or form outside the target EFE objective.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim depends on the validity of the chosen priors and the assumption that the generative model structure allows the decomposition.

axioms (1)
  • domain assumption Planning can be cast as inference under the Free Energy Principle
    Invoked to motivate the unification of planning with perception and learning.
invented entities (1)
  • epistemic priors no independent evidence
    purpose: To augment the generative model so that VFE minimization yields the EFE decomposition
    These priors are introduced in the paper to achieve the desired objective decomposition.

pith-pipeline@v0.9.1-grok · 5732 in / 1123 out tokens · 27964 ms · 2026-06-27T13:37:36.812591+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

300 extracted references · 180 canonical work pages · 21 internal anchors

  1. [1]

    and Shipp, Stewart and Friston, Karl J

    Adams, Rick A. and Shipp, Stewart and Friston, Karl J. , year = 2013, journal =. Predictions Not Commands: Active Inference in the Motor System , shorttitle =. doi:10.1007/s00429-012-0475-5 , urldate =

    work page doi:10.1007/s00429-012-0475-5 2013

  2. [2]

    and Murphy, Kevin , year = 2018, month = jul, pages =

    Alemi, Alexander and Poole, Ben and Fischer, Ian and Dillon, Joshua and Saurous, Rif A. and Murphy, Kevin , year = 2018, month = jul, pages =. Fixing a. Proceedings of the 35th

    2018

  3. [3]

    Computationally Efficient Convolved Multiple Output

    Alvarez, Mauricio A and Lawrence, Neil D , year = 2011, journal =. Computationally Efficient Convolved Multiple Output

    2011

  4. [4]

    and Rosasco, Lorenzo and Lawrence, Neil D

    Alvarez, Mauricio A. and Rosasco, Lorenzo and Lawrence, Neil D. , year = 2012, month = apr, number =. Kernels for. arXiv , langid =:1106.6251 , primaryclass =

    Pith/arXiv arXiv 2012

  5. [5]

    , year = 2012, month = jun, pages =

    Anandkumar, Animashree and Hsu, Daniel and Kakade, Sham M. , year = 2012, month = jun, pages =. A. Proceedings of the 25th

    2012

  6. [6]

    Anil Meera, Ajith and Lanillos, Pablo , editor =. Towards. Active. doi:10.1007/978-3-031-47958-8_3 , abstract =

    work page doi:10.1007/978-3-031-47958-8_3

  7. [7]

    Biosystems , volume =

    Natural Hierarchy Emerges from Energy Dispersal , author =. Biosystems , volume =. doi:10.1016/j.biosystems.2008.10.008 , urldate =

    work page doi:10.1016/j.biosystems.2008.10.008 2008

  8. [8]

    Declarative

    Antoy, Sergio and Hanus, Michael , editor =. Declarative. Logic. doi:10.1007/11680093_2 , abstract =

    work page doi:10.1007/11680093_2

  9. [9]

    Planning by Probabilistic Inference , booktitle =

    Attias, Hagai , year = 2003, pages =. Planning by Probabilistic Inference , booktitle =

    2003

  10. [10]

    Reinforcement

    Azizzadenesheli, Kamyar and Lazaric, Alessandro and Anandkumar, Animashree , year = 2016, month = jun, pages =. Reinforcement. Conference on

    2016

  11. [11]

    Reactive

    Bagaev, Dmitry and. Reactive. doi:10.48550/arXiv.2112.13251 , urldate =. arXiv , keywords =:2112.13251 , primaryclass =

    work page doi:10.48550/arxiv.2112.13251

  12. [12]

    Reactive

    Bagaev, Dmitry and De Vries, Bert , editor =. Reactive. Scientific Programming , volume =. doi:10.1155/2023/6601690 , urldate =

    work page doi:10.1155/2023/6601690 2023

  13. [13]

    Software Impacts , volume =

    Bagaev, Dmitry and. Software Impacts , volume =. doi:10.1016/j.simpa.2022.100299 , urldate =

    work page doi:10.1016/j.simpa.2022.100299 2022

  14. [14]

    doi:10.21105/joss.05161 , urldate =

    Bagaev, Dmitry and Podusenko, Albert and de Vries, Bert , year = 2023, month = apr, journal =. doi:10.21105/joss.05161 , urldate =

    work page doi:10.21105/joss.05161 2023

  15. [15]

    ACM Computing Surveys (CSUR) , volume =

    A Survey on Reactive Programming , author =. ACM Computing Surveys (CSUR) , volume =. doi:10.1145/2501654.2501666 , urldate =

    work page doi:10.1145/2501654.2501666

  16. [16]

    and Simpson, Daniel and Rue, H

    Bakka, Haakon and Vanhatalo, Jarno and Illian, Janine B. and Simpson, Daniel and Rue, H. Non-Stationary. Spatial Statistics , volume =. doi:10.1016/j.spasta.2019.01.002 , urldate =

    work page doi:10.1016/j.spasta.2019.01.002 2019

  17. [17]

    and Daulton, Samuel and Letham, Benjamin and Wilson, Andrew Gordon and Bakshy, Eytan , year = 2020, month = dec, series =

    Balandat, Maximilian and Karrer, Brian and Jiang, Daniel R. and Daulton, Samuel and Letham, Benjamin and Wilson, Andrew Gordon and Bakshy, Eytan , year = 2020, month = dec, series =. Proceedings of the 34th

    2020

  18. [18]

    arXiv , keywords =:2111.10530 , primaryclass =

    Kalman Filters as the Steady-State Solution of Gradient Descent on Variational Free Energy , author =. arXiv , keywords =:2111.10530 , primaryclass =

    arXiv

  19. [19]

    Reactive Probabilistic Programming , booktitle =

    Baudart, Guillaume and Mandel, Louis and Atkinson, Eric and Sherman, Benjamin and Pouzet, Marc and Carbin, Michael , year = 2020, month = jun, series =. Reactive Probabilistic Programming , booktitle =. doi:10.1145/3385412.3386009 , urldate =

    work page doi:10.1145/3385412.3386009 2020

  20. [20]

    Beck, Jeff and Ramstead, Maxwell J. D. , year = 2025, month = feb, number =. Dynamic. doi:10.48550/arXiv.2502.21217 , urldate =. arXiv , keywords =:2502.21217 , primaryclass =

    work page doi:10.48550/arxiv.2502.21217 2025

  21. [21]

    Bellman, Richard , year = 1966, journal =. Dynamic. 1719695 , eprinttype =

    1966

  22. [22]

    Bulletin of the American Mathematical Society , volume =

    The Theory of Dynamic Programming , author =. Bulletin of the American Mathematical Society , volume =. doi:10.1090/S0002-9904-1954-09848-8 , urldate =

    work page doi:10.1090/s0002-9904-1954-09848-8 1954

  23. [23]

    Robustness in Identification and Control , author =

    Robust Model Predictive Control:. Robustness in Identification and Control , author =. doi:10.1007/BFb0109870 , abstract =

    work page doi:10.1007/bfb0109870

  24. [24]

    Multidimensional Binary Search Trees Used for Associative Searching , author =. Commun. ACM , volume =. doi:10.1145/361002.361007 , urldate =

    work page doi:10.1145/361002.361007

  25. [25]

    Dynamic Programming and Optimal Control:

    Bertsekas, Dimitri , year = 2012, volume =. Dynamic Programming and Optimal Control:

    2012

  26. [26]

    Julia: A Fresh Approach to Numerical Computing

    Bezanson, Jeff and Edelman, Alan and Karpinski, Stefan and Shah, Viral B. , year = 2015, month = jul, number =. Julia:. doi:10.48550/arXiv.1411.1607 , urldate =. arXiv , keywords =:1411.1607 , primaryclass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1411.1607 2015

  27. [27]

    Bingham, Eli and Chen, Jonathan P and Jankowiak, Martin and Obermeyer, Fritz and Pradhan, Neeraj and Karaletsos, Theofanis and Singh, Rohit and Szerlip, Paul and Horsfall, Paul and Goodman, Noah D , year = 2019, journal =. Pyro:

    2019

  28. [28]

    Pattern Recognition and Machine Learning , author =

  29. [29]

    , year = 2011, month = dec, journal =

    Blackmore, Lars and Ono, Masahiro and Williams, Brian C. , year = 2011, month = dec, journal =. Chance-. doi:10.1109/TRO.2011.2161160 , urldate =

    work page doi:10.1109/tro.2011.2161160 2011

  30. [30]

    and Jordan, Michael I

    Blei, David M. and Jordan, Michael I. , year = 2006, month = mar, journal =. Variational Inference for. doi:10.1214/06-BA104 , urldate =

    work page doi:10.1214/06-ba104 2006

  31. [31]

    Journal of the American Statistical Association , author =

    Blei, David M. and Kucukelbir, Alp and McAuliffe, Jon D. , year = 2017, month = apr, journal =. Variational. doi:10.1080/01621459.2017.1285773 , urldate =

    work page doi:10.1080/01621459.2017.1285773 2017

  32. [32]

    Bayesian

    Bliznyuk, Nikolay and Ruppert, David and Shoemaker, Christine and Regis, Rommel and Wild, Stefan and Mugunthan, Pradeep , year = 2008, month = jun, journal =. Bayesian. doi:10.1198/106186008X320681 , urldate =

    work page doi:10.1198/106186008x320681 2008

  33. [33]

    Biosystems Engineering , volume =

    Minimising the Non-Working Distance Travelled by Machines Operating in a Headland Field Pattern , author =. Biosystems Engineering , volume =. doi:10.1016/j.biosystemseng.2008.06.008 , urldate =

    work page doi:10.1016/j.biosystemseng.2008.06.008 2008

  34. [34]

    Bolin, David and Lindgren, Finn , year = 2011, journal =. Spatial. 23024839 , eprinttype =

    2011

  35. [35]

    Artificial Intelligence , volume =

    Planning as Heuristic Search , author =. Artificial Intelligence , volume =. doi:10.1016/S0004-3702(01)00108-4 , urldate =

    work page doi:10.1016/s0004-3702(01)00108-4

  36. [36]

    Multi-Task

    Bonilla, Edwin V and Chai, Kian and Williams, Christopher , year = 2007, volume =. Multi-Task. Advances in

    2007

  37. [37]

    The International Journal of Robotics Research , volume =

    Closing the Learning-Planning Loop with Predictive State Representations , author =. The International Journal of Robotics Research , volume =. doi:10.1177/0278364911404092 , urldate =

    work page doi:10.1177/0278364911404092

  38. [38]

    Mat\'ern

    Borovitskiy, Viacheslav and Terenin, Alexander and Mostowsky, Peter and Deisenroth, Marc Peter , year = 2020, month = dec, series =. Mat\'ern. Proceedings of the 34th

    2020

  39. [39]

    Trends in Cognitive Sciences , volume =

    Planning as Inference , author =. Trends in Cognitive Sciences , volume =. doi:10.1016/j.tics.2012.08.006 , urldate =

    work page doi:10.1016/j.tics.2012.08.006 2012

  40. [40]

    Generating Sentences from a Continuous Space , booktitle =

    Bowman, Samuel and Vilnis, Luke and Vinyals, Oriol and Dai, Andrew and Jozefowicz, Rafal and Bengio, Samy , year = 2016, pages =. Generating Sentences from a Continuous Space , booktitle =

    2016

  41. [41]

    Bradbury, James and Frostig, Roy and Hawkins, Peter and Johnson, Matthew James and Leary, Chris and Maclaurin, Dougal and Necula, George and Paszke, Adam and VanderPlas, Jake and

  42. [42]

    OpenAI Gym

    Brockman, Greg and Cheung, Vicki and Pettersson, Ludwig and Schneider, Jonas and Schulman, John and Tang, Jie and Zaremba, Wojciech , year = 2016, month = jun, number =. doi:10.48550/arXiv.1606.01540 , urldate =. arXiv , keywords =:1606.01540 , primaryclass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1606.01540 2016

  43. [43]

    Scalable

    Bruinsma, Wessel and Perim, Eric and Tebbutt, William and Hosking, Scott and Solin, Arno and Turner, Richard , year = 2020, month = nov, pages =. Scalable. Proceedings of the 37th

    2020

  44. [44]

    and Kim, Chang Sub and McGregor, Simon and Seth, Anil K

    Buckley, Christopher L. and Kim, Chang Sub and McGregor, Simon and Seth, Anil K. , year = 2017, month = dec, journal =. The Free Energy Principle for Action and Perception:. doi:10.1016/j.jmp.2017.09.004 , urldate =

    work page doi:10.1016/j.jmp.2017.09.004 2017

  45. [45]

    and Nguyen, Cuong V

    Bui, Thang D. and Nguyen, Cuong V. and Turner, Richard E. , year = 2017, month = dec, series =. Streaming Sparse. Proceedings of the 31st

    2017

  46. [46]

    Finding the Outliers in Scanpath Data , booktitle =

    Burch, Michael and Kumar, Ayush and Mueller, Klaus and Kervezee, Titus and Nuijten, Wouter and Oostenbach, Rens and Peeters, Lucas and Smit, Gijs , year = 2019, month = jun, series =. Finding the Outliers in Scanpath Data , booktitle =. doi:10.1145/3317958.3318225 , urldate =

    work page doi:10.1145/3317958.3318225 2019

  47. [47]

    and Lee, Daniel and Goodrich, Ben and Betancourt, Michael and Brubaker, Marcus and Guo, Jiqiang and Li, Peter and Riddell, Allen , year = 2017, journal =

    Carpenter, Bob and Gelman, Andrew and Hoffman, Matthew D. and Lee, Daniel and Goodrich, Ben and Betancourt, Michael and Brubaker, Marcus and Guo, Jiqiang and Li, Peter and Riddell, Allen , year = 2017, journal =. doi:10.18637/jss.v076.i01 , urldate =

    work page doi:10.18637/jss.v076.i01 2017

  48. [48]

    Stochastic Versions of the Em Algorithm: An Experimental Study in the Mixture Case , shorttitle =

    Celeux,. Stochastic Versions of the Em Algorithm: An Experimental Study in the Mixture Case , shorttitle =. Journal of Statistical Computation and Simulation , volume =. doi:10.1080/00949659608811772 , urldate =

    work page doi:10.1080/00949659608811772

  49. [49]

    Branching

    Champion, Th. Branching. Neural Networks , volume =. doi:10.1016/j.neunet.2022.03.036 , urldate =

    work page doi:10.1016/j.neunet.2022.03.036 2022

  50. [50]

    Chang, Huiwen and Zhang, Han and Barber, Jarred and Maschinot, A. J. and Lezama, Jose and Jiang, Lu and Yang, Ming-Hsuan and Murphy, Kevin and Freeman, William T. and Rubinstein, Michael and Li, Yuanzhen and Krishnan, Dilip , year = 2023, month = jan, number =. Muse:. doi:10.48550/arXiv.2301.00704 , urldate =. arXiv , keywords =:2301.00704 , primaryclass =

    work page doi:10.48550/arxiv.2301.00704 2023

  51. [51]

    Chen, Fan and Wang, Huan and Xiong, Caiming and Mei, Song and Bai, Yu , year = 2023, month = jul, pages =. Lower. Proceedings of the 40th

    2023

  52. [52]

    Physical Review E , volume =

    Loop Calculus in Statistical Physics and Information Science , author =. Physical Review E , volume =. doi:10.1103/PhysRevE.73.065102 , urldate =

    work page doi:10.1103/physreve.73.065102

  53. [53]

    Advances in Neural Information Processing Systems , volume =

    Minigrid & Miniworld:. Advances in Neural Information Processing Systems , volume =

  54. [54]

    Temporal

    Corenflos, Adrien and Zhao, Zheng and S. Temporal. 2022 25th. doi:10.23919/FUSION49751.2022.9841306 , urldate =

    work page doi:10.23919/fusion49751.2022.9841306 2022

  55. [55]

    International

    Cox, Marco and. International

  56. [56]

    American journal of physics , volume =

    Probability, Frequency and Reasonable Expectation , author =. American journal of physics , volume =

  57. [57]

    and Ramaker, B

    Cutler, Richard R. and Ramaker, B. L. , year = 1979, journal =. Dynamic

    1979

  58. [58]

    Active Inference on Discrete State-Spaces:

    Da Costa, Lancelot and Parr, Thomas and Sajid, Noor and Veselic, Sebastijan and Neacsu, Victorita and Friston, Karl , year = 2020, month = dec, journal =. Active Inference on Discrete State-Spaces:. doi:10.1016/j.jmp.2020.102447 , urldate =

    work page doi:10.1016/j.jmp.2020.102447 2020

  59. [59]

    Da Costa, Lancelot and Tenka, Samuel and Zhao, Dominic and Sajid, Noor , year = 2024, month = jan, number =. Active. doi:10.48550/arXiv.2401.12917 , urldate =. arXiv , keywords =:2401.12917 , primaryclass =

    work page doi:10.48550/arxiv.2401.12917 2024

  60. [60]

    doi:10.5194/essd-15-317-2023 , urldate =

    Earth System Science Data , volume =. doi:10.5194/essd-15-317-2023 , urldate =

    work page doi:10.5194/essd-15-317-2023 2023

  61. [61]

    IEEE Transactions on Geoscience and Remote Sensing , volume =

    Bayesian. IEEE Transactions on Geoscience and Remote Sensing , volume =. doi:10.1109/TGRS.2024.3434443 , urldate =

    work page doi:10.1109/tgrs.2024.3434443 2024

  62. [62]

    Differentiable Expected Hypervolume Improvement for Parallel Multi-Objective

    Daulton, Samuel and Balandat, Maximilian and Bakshy, Eytan , year = 2020, month = dec, series =. Differentiable Expected Hypervolume Improvement for Parallel Multi-Objective. Proceedings of the 34th

    2020

  63. [63]

    Parallel

    Daulton, Samuel and Balandat, Maximilian and Bakshy, Eytan , year = 2021, month = dec, series =. Parallel. Proceedings of the 35th

    2021

  64. [64]

    , year = 2007, month = jun, pages =

    Dauwels, J. , year = 2007, month = jun, pages =. On. doi:10.1109/ISIT.2007.4557602 , abstract =

    work page doi:10.1109/isit.2007.4557602 2007

  65. [65]

    doi:10.1109/MMAR.2012.6347921 , urldate =

    Consistent Control Hierarchies with Top Layers Represented by Timed Event Graphs , booktitle =. doi:10.1109/MMAR.2012.6347921 , urldate =

    work page doi:10.1109/mmar.2012.6347921 2012

  66. [66]

    Entropy , volume =

    Spatial and. Entropy , volume =. doi:10.3390/e26010083 , urldate =

    work page doi:10.3390/e26010083

  67. [67]

    Expected

    De Vries, Bert and Nuijten, Wouter and. Expected. doi:10.48550/arXiv.2504.14898 , urldate =. arXiv , keywords =:2504.14898 , primaryclass =

    work page doi:10.48550/arxiv.2504.14898

  68. [68]

    Journal of neural engineering , volume =

    The Neural Optimal Control Hierarchy for Motor Control , author =. Journal of neural engineering , volume =

  69. [69]

    TensorFlow Distributions

    Dillon, Joshua V. and Langmore, Ian and Tran, Dustin and Brevdo, Eugene and Vasudevan, Srinivas and Moore, Dave and Patton, Brian and Alemi, Alex and Hoffman, Matt and Saurous, Rif A. , year = 2017, month = nov, number =. doi:10.48550/arXiv.1711.10604 , urldate =. arXiv , keywords =:1711.10604 , primaryclass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1711.10604 2017

  70. [70]

    Stochastic

    Dritsas, Ioannis , year = 2011, month = feb, publisher =. Stochastic

    2011

  71. [71]

    and Del Bello, U

    Drusch, M. and Del Bello, U. and Carlier, S. and Colin, O. and Fernandez, V. and Gascon, F. and Hoersch, B. and Isola, C. and Laberinti, P. and Martimort, P. and Meygret, A. and Spoto, F. and Sy, O. and Marchese, F. and Bargellini, P. , year = 2012, month = may, journal =. Sentinel-2:. doi:10.1016/j.rse.2011.11.026 , urldate =

    work page doi:10.1016/j.rse.2011.11.026 2012

  72. [72]

    Duane, Simon and Kennedy, A. D. and Pendleton, Brian J. and Roweth, Duncan , year = 1987, month = sep, journal =. Hybrid. doi:10.1016/0370-2693(87)91197-X , urldate =

    work page doi:10.1016/0370-2693(87)91197-x 1987

  73. [73]

    Durrande, Nicolas and Adam, Vincent and Bordeaux, Lucas and Eleftheriadis, Stefanos and Hensman, James , year = 2019, month = apr, pages =. Banded. Proceedings of the

    2019

  74. [74]

    and Choo, Xuan and Bekolay, Trevor and DeWolf, Travis and Tang, Yichuan and Rasmussen, Daniel , year = 2012, month = nov, journal =

    Eliasmith, Chris and Stewart, Terrence C. and Choo, Xuan and Bekolay, Trevor and DeWolf, Travis and Tang, Yichuan and Rasmussen, Daniel , year = 2012, month = nov, journal =. A. doi:10.1126/science.1225266 , urldate =

    work page doi:10.1126/science.1225266 2012

  75. [75]

    Resolving Uncertainty on the Fly:

    Engstr. Resolving Uncertainty on the Fly:. doi:10.48550/arXiv.2311.06417 , urldate =. arXiv , keywords =:2311.06417 , primaryclass =

    work page doi:10.48550/arxiv.2311.06417

  76. [76]

    Scalable Global Optimization via Local

    Eriksson, David and Pearce, Michael and Gardner, Jacob R and Turner, Ryan and Poloczek, Matthias , year = 2019, month = dec, number =. Scalable Global Optimization via Local. Proceedings of the 33rd

    2019

  77. [77]

    Dynamical

    Esaki, Kanako and Matsumura, Tadayuki and Minusa, Shunsuke and Shao, Yang and Yoshimura, Chihiro and Mizuno, Hiroyuki , editor =. Dynamical. Active. doi:10.1007/978-3-031-47958-8_2 , abstract =

    work page doi:10.1007/978-3-031-47958-8_2

  78. [78]

    Hierarchical

    Findeisen, W. Hierarchical

  79. [79]

    Annealing in Variational Inference Mitigates Mode Collapse:

    Fogliani, Luigi and Loureiro, Bruno and Gabri. Annealing in Variational Inference Mitigates Mode Collapse:. doi:10.48550/arXiv.2602.12923 , urldate =. arXiv , keywords =:2602.12923 , primaryclass =

    work page doi:10.48550/arxiv.2602.12923

  80. [80]

    David , year = 2001, journal =

    Forney, G. David , year = 2001, journal =. Codes on Graphs:

    2001

Showing first 80 references.