Pith. sign in

REVIEW 2 major objections 2 minor 92 references

Internal-state criticality in Bayesian-inverse-Bayesian inference

T0 review · 2 major / 2 minor · reviewed 2026-06-26 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Adding an inverse-Bayesian relaxation step renders Bayesian inference critical without external parameter adjustment.

desk verdict BIB adds one inverse relaxation step to Bayesian updating and gets robust criticality in cyclic games via boundary reconstruction, shown through consistent numerics and controls. read the letter →

arxiv 2606.22109 v2 pith:Z4ZD4LFM submitted 2026-06-20 cond-mat.stat-mech nlin.AO

Kazuto Sasai , Yukio-Pegio Gunji This is my paper · ORCID
classification cond-mat.stat-mechnlin.AO
keywords Bayesian-inverse-Bayesianinferencecriticalityrock-paper-scissorsheavy-taileddistributionsrepeatedgameshypothesisrenewalstatisticalmechanics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper establishes that Bayesian-inverse-Bayesian inference, which incorporates an inverse step of hypothesis renewal, generates critical dynamics in repeated games purely from the inference mechanism. In N-hand rock-paper-scissors simulations where other algorithms reduce to random play, BIB produces a heavy-tailed power-law distribution in argmax persistence with exponent near 1.43, robust across conditions. This criticality manifests as a zero-drift state in the log-posterior walk and shows data collapse and finite-size scaling. Ordinary Bayesian inference without the inverse step exhibits no such universality or power laws. The mechanism works by continually reconstructing the hypothesis-space boundary rather than converging to an absorbing state.

What carries the argument

The inverse-Bayesian relaxation step of hypothesis renewal that continually reconstructs the hypothesis-space boundary in BIB inference.

What would settle it

If the argmax-persistence distribution in BIB simulations fails to follow a heavy-tailed power law or to exhibit cross-design data collapse while Bayes-only inference remains non-critical, the claim of internal-state criticality would be falsified.

Watch

Extended reading notes

Core claim

In the BIB model applied to repeated cyclic-dominance games, the dynamics remain in an internal critical state with the argmax-persistence distribution following a heavy-tailed power law, invariant across hypothesis counts and showing finite-size scaling, while the exponent drifts toward 3/2 with vanishing finite-sample residual; this state is reached by the inverse-Bayesian relaxation step that reconstructs the hypothesis boundary, and is absent in Bayes-only inference.

Load-bearing premise

The N-hand cyclic-dominance rock-paper-scissors simulation with Nash-targeting algorithms collapsing to uniform random play is a faithful minimal setting in which any non-trivial dynamics must originate internally from the inference rule itself.

Editorial extensions

If this is right

  • The argmax-persistence distribution stays a heavy-tailed power law with exponent approximately 1.43 at the canonical window.
  • Along window and alphabet axes the exponent drifts toward the universal 3/2 as the finite-sample residual (N-1)/(2m) vanishes.
  • The critical state remains invariant across the hypothesis count Nh, with cutoff time and posterior spread obeying finite-size scaling.
  • Cross-design data collapse occurs for the zero-drift log-posterior walk, confirming robustness rather than any single exponent value.
  • Bayes-only inference shows no analogous universality and no power law in the same setting.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The boundary-reconstruction route to criticality could extend to other repeated-decision settings where agents must avoid absorbing states without external tuning.
  • The first-passage reading of argmax and laminar observables as a driftless walk suggests direct connections to other models of persistent fluctuations in statistical mechanics.
  • Testing BIB on continuous action spaces or non-cyclic games would clarify whether the internal criticality persists beyond the discrete minimal case.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes Bayesian-inverse-Bayesian (BIB) inference as a minimal generative model for critical dynamics in repeated games. In N-hand cyclic-dominance rock-paper-scissors simulations, where Nash-targeting collapses to uniform play, the addition of an inverse-Bayesian relaxation (hypothesis renewal) step produces heavy-tailed argmax-persistence distributions with power-law exponent α≈1.43 at the canonical window size; the exponent drifts toward the universal 3/2 value as the finite-sample residual (N−1)/(2m) vanishes. Robustness is evidenced by cross-design data collapse onto a zero-drift log-posterior walk, finite-size scaling with hypothesis count Nh, and explicit contrast to Bayes-only inference, which lacks both the power law and universality. The central claim is that the inverse step alone renders the dynamics critical by continually reconstructing the hypothesis-space boundary, without external parameter adjustment.

Significance. If substantiated by the full derivations and numerics, the result supplies a concrete, minimal link between Bayesian inference procedures and critical phenomena in statistical mechanics, with the data-collapse and finite-size scaling evidence providing a stronger test of universality than any single exponent. The mechanism of boundary reconstruction offers a complementary route to criticality that is robust across a natural parameter range rather than tuned to an absorbing state, which could inform models of heavy-tailed statistics in inference-driven systems.

major comments (2)
  1. [Simulation setup] The isolation of non-trivial dynamics to the inference rule itself rests on the claim that Nash-targeting algorithms collapse to uniform random play. Without a quantitative demonstration (e.g., measured bias or residual correlation in the opponent strategy distribution) in the simulation setup section, it remains possible that weak residual structure in the game dynamics contributes to the observed power laws.
  2. [Abstract and §4] The abstract asserts that the state is 'parameter-free' and reached 'with no external parameter adjustment,' yet identifies a canonical window size m whose residual (N−1)/(2m) controls the drift of the exponent toward 3/2. Clarification is required on whether m is an externally chosen scale or emerges internally, as this directly affects the load-bearing claim of robustness without tuning.
minor comments (2)
  1. [Abstract] The abstract reports simulation results and data collapse but supplies no equations for the log-posterior walk or error bars on the fitted exponents; the full manuscript should include these in the methods or results section for reproducibility.
  2. [Results] Notation for the hypothesis count Nh versus the number of hands N should be introduced once and used consistently to avoid confusion in the finite-size scaling discussion.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment and recommendation of minor revision. The comments identify points where additional evidence and clarification will strengthen the manuscript, which we address point by point below.

read point-by-point responses

  1. Referee: [Simulation setup] The isolation of non-trivial dynamics to the inference rule itself rests on the claim that Nash-targeting algorithms collapse to uniform random play. Without a quantitative demonstration (e.g., measured bias or residual correlation in the opponent strategy distribution) in the simulation setup section, it remains possible that weak residual structure in the game dynamics contributes to the observed power laws.

    Authors: We agree that explicit quantitative support would strengthen the claim. The manuscript invokes the known result that the unique Nash equilibrium of cyclic N-hand rock-paper-scissors is the uniform mixed strategy, so that Nash-targeting play converges to uniform random. To address the concern directly, the revised simulation-setup section will include measured bias and lagged autocorrelation of the opponent strategy distribution, confirming that residuals remain at the level of finite-sample noise with no detectable structure capable of generating the reported power laws. revision: yes

  2. Referee: [Abstract and §4] The abstract asserts that the state is 'parameter-free' and reached 'with no external parameter adjustment,' yet identifies a canonical window size m whose residual (N−1)/(2m) controls the drift of the exponent toward 3/2. Clarification is required on whether m is an externally chosen scale or emerges internally, as this directly affects the load-bearing claim of robustness without tuning.

    Authors: The window size m is an externally chosen design parameter of the inference update. The phrase 'no external parameter adjustment' is intended to convey that criticality does not require tuning m (or any other hyperparameter) to a special value; the zero-drift log-posterior walk and cross-design data collapse remain robust for any finite m, with the exponent approaching the universal 3/2 only in the vanishing-residual limit. The inverse-Bayesian renewal step itself produces the critical state by boundary reconstruction for generic m. The revised abstract and §4 will state this distinction explicitly. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity identified

full rationale

The paper's central claim—that adding the inverse-Bayesian relaxation step produces a robust critical state via boundary reconstruction—is supported by explicit simulation contrasts (BIB vs. Bayes-only), multi-axis parameter sweeps, data collapse onto zero-drift behavior, and finite-size scaling with Nh. These are empirical observations from the RPS setup, not a mathematical derivation that reduces the reported power-law or zero-drift property to a fitted input or self-defined quantity by construction. The abstract explicitly frames robustness as the collapse itself rather than any fixed exponent, and no load-bearing step equates a prediction to its own definition or relies on an unverified self-citation chain. The derivation chain remains self-contained against the described external benchmarks.

Assumptions & free parameters 1 free parameters · 1 assumptions · 1 invented entities

The central claim rests on the assumption that the discrete RPS simulation isolates internal dynamics and that power-law persistence plus data collapse constitute evidence of a critical zero-drift state. No free parameters are explicitly fitted in the abstract, but the canonical window and the reported exponent 1.43 imply choices. The inverse-Bayesian relaxation step is an invented modeling construct whose independent evidence is the simulation itself.

free parameters (1)
  • canonical window size m
    Used as reference point for the reported exponent; sweeps are claimed to show robustness but the specific value is selected.
assumptions (1)
  • domain assumption The argmax and laminar observables are first-passage reads of one driftless log-posterior walk.
    Invoked to interpret the power-law persistence as evidence of a critical zero-drift state.
invented entities (1)
  • Bayesian-inverse-Bayesian (BIB) inference procedure
    purpose: Generative model that adds hypothesis renewal to standard Bayesian updating to produce criticality
    The inverse step is introduced by the authors; no external falsifiable prediction outside the simulation is stated in the abstract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Internal-state criticality in Bayesian-inverse-Bayesian inference." pith.science (2026). https://pith.science/paper/Z4ZD4LFM

@misc{pith2026260622109,
  author       = {Pith},
  title        = {Pith review of: Internal-state criticality in Bayesian-inverse-Bayesian inference},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z4ZD4LFM}},
  note         = {Machine review of arXiv:2606.22109}
}
abstract

We propose Bayesian-inverse-Bayesian (BIB) inference in repeated games as a minimal generative model linking Bayesian inference, statistical mechanics, and heavy-tailed statistics. As a concrete instantiation we simulate repeated $N$-hand cyclic-dominance rock-paper-scissors, a discrete setting in which Nash-targeting algorithms collapse to uniform random play, so that any non-trivial dynamics must originate internally. Across a multi-axis sweep of design, window, and opponent conditions, the BIB dynamics remain in the same internal critical state, the argmax-persistence distribution staying a heavy-tailed power law with exponent $\alpha\approx 1.43$ at the canonical window. Along the window and alphabet axes the exponent is not constant but drifts toward the universal $3/2$ as the finite-sample residual $(N-1)/(2m)$ vanishes. Bayes-only inference, which lacks the inverse step, shows no analogous universality and no power law. Because the argmax and laminar observables are first-passage reads of one driftless log-posterior walk, what is robust across conditions is the critical, zero-drift state itself, evidenced by the cross-design data collapse rather than by any particular exponent value. The state is also invariant across the hypothesis count $N_h$, with the cutoff time and posterior spread obeying finite-size scaling. Adding an inverse-Bayesian relaxation step (hypothesis renewal) to ordinary Bayesian inference is by itself enough to render the dynamics critical, with no external parameter adjustment. Rather than self-organizing toward an absorbing state, BIB reaches criticality by continually reconstructing the hypothesis-space boundary, a mechanism complementary to self-organized criticality that makes the criticality robust across a natural parameter range.

Figures

Figures reproduced from arXiv: 2606.22109 by the authors.

Figure 1
Figure 1. FIG. 1. Schematic of the Bayesian and inverse-Bayesian steps and the resulting internal-state critical dynamics. The Bayesian [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Hypothesis-space dynamics, BIB-BIB (left column) versus BO-BO (right column). Same random seed, rs design [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Design-independent collapse of the BIB argmax [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Specificity of the critical class to the inverse-Bayesian rule among adaptive learners. Internal argmax-persistence [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Extrapolation of the argmax exponent to 3 [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. The [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. BIB versus BO, head-to-head reward and cross-design tournament (at [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8. Internal-state criticality versus behavioral expression. Complementary cumulative distributions (CCDFs), BIB–BIB, [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: FIG. 9. The predicted behavioral crossover in existing human play. Re-analysis of the public human-versus-bot data of [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: FIG. 10 [PITH_FULL_IMAGE:figures/full_fig_p019_10.png]
Figure 11
Figure 11. Figure 11: FIG. 11. The drift as the relevant variable, with [PITH_FULL_IMAGE:figures/full_fig_p022_11.png]
Figure 12
Figure 12. Figure 12: FIG. 12. BIB robustness sweeps along two control axes. (a) Window-size dependence, pooled BIB-BIB argmax-persistence [PITH_FULL_IMAGE:figures/full_fig_p023_12.png]
Figure 13
Figure 13. Figure 13: FIG. 13. Observation-rule ablation at medium scale ( [PITH_FULL_IMAGE:figures/full_fig_p025_13.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

92 extracted references · 4 canonical work pages

  1. [1]

    Each implementation dif- fers only in the posterior-regularization rule (Table II)

    Comparison design We ran a controlled comparison at the canonical con- dition (rs design, BIB-BIB,N h = 10,m= 50,T= 2000 steps with the last 1000 analyzed, 100 runs, identical seeds across implementations). Each implementation dif- fers only in the posterior-regularization rule (Table II)

    2000

  2. [2]

    Dynamics of the four rules The three conditional rules are statistically indis- tinguishable in their dynamics (¯σ≈0.13, argmax- persistenceα≈1.36–1.38, mean argmax run length≈34 steps, similar inter-run SD). The always-on rule shows an∼8-fold collapse in ¯σ, an∼8-fold increase in argmax switches, a collapse of the mean argmax run length from ≈34 to≈4 ste...

  3. [3]

    Why conditional triggering matters The conditional structure itself is essential. The con- ditional trigger acts as arelaxation mechanismthat fires only when the Bayesian concentration brings the sys- tem to the edge of collapse, mirroring the on-off inter- mittency mechanism of Platt, Spiegel and Tresser [63] in which a slow driving variable intermittent...

  4. [4]

    After discarding the first T /2 = 10 5 steps as burn-in, the remaining 10 5 ×20 = 2×10 6 decision events are pooled for statistical analysis

    Simulation scale and runs The main results useT= 2×10 5 steps and 20 in- dependent runs per condition. After discarding the first T /2 = 10 5 steps as burn-in, the remaining 10 5 ×20 = 2×10 6 decision events are pooled for statistical analysis. For the central design×window×pair sweep (96 condi- tions) this yields∼1.9×10 8 decision events in total. The Nh...

  5. [5]

    Long persistence in- dicates the agent continuously trusts a single dominant hypothesis

    Metrics The principal observable is theargmax-hypothesis per- sistence timeT argmax, defined as the run-length of con- secutive time stepston which the argmax indexi ∗(t) := arg maxi Pt(hi) remains unchanged. Long persistence in- dicates the agent continuously trusts a single dominant hypothesis. Short persistence indicates rapid hypothe- sis switching. F...

  6. [6]

    Implementation and reproducibility All simulations are implemented in Python (NumPy and multiprocessing for parallelization,powerlaw for fitting, pandas for aggregation). The full source code, simulation scripts, JSON-serialized intermediate results, and aggregate CSVs are pub- licly available athttps://github.com/kazsasai/ bayesian-inverse-bayesian-rpsan...

  7. [7]

    Model selection for the persistence-time distributions Throughout, we report a power-law exponent only for conditions that pass a fixed heavy-tail criterion. For each pooled persistence-time set we select the lower cut- offx min by the Clauset–Shalizi–Newman procedure [56] and compare five candidate distributions, namely power law, truncated power law, ex...

  8. [8]

    III A, the maximal argmax-stable intervals in whichP(h ∗)> θ= 0.3 for a fractionf= 0.8 of the run, gives a stricter test of the collapse

    Plateau-length distribution The plateau observable of Sec. III A, the maximal argmax-stable intervals in whichP(h ∗)> θ= 0.3 for a fractionf= 0.8 of the run, gives a stricter test of the collapse. The BIB-BIB plateau-length CCDF passes the heavy-tail criterion of Appendix B 5 (about 3.6 decades, gap 0.18,α≈1.23), whereas BO-BO does not

Show all 92 references
  1. [9]

    IV A places the argmax persistence in the on-off intermittency class through the first-return statistics of the log-posterior walk

    On-off intermittency of the log-posterior walk: supporting analysis The two-hypothesis reduction of Sec. IV A places the argmax persistence in the on-off intermittency class through the first-return statistics of the log-posterior walk. This appendix collects the supporting an...

  2. [10]

    V B), we applied the identical inference core to matching pennies, a two-action zero- sum game with a uniform Nash equilibrium (1/2,1/2) and no cyclic dominance

    Matching-pennies control To test whether the BIB critical class is intrinsic to the inverse-Bayesian rule rather than to the cyclic-dominance structure of RPS (Sec. V B), we applied the identical inference core to matching pennies, a two-action zero- sum game with a uniform Na...

  3. [11]

    Specificity among reinforcement-learning baselines To test whether the argmax-persistence (α≈1.43) and laminar (α≈1.325) criticality is specific to BIB or generic to adaptive play on these uniform-Nash games (Sec. III B), we compared BIB with three standard, parameter-light le...

  4. [12]

    Behavioral re-analysis of existing human play This appendix gives the data and methods behind Sec. VII B. The behavioral corollary predicts that a player’s behavioral persistence is short-tailed against an adaptive opponent, which pins the action stream toward the uniform Nash...

  5. [13]

    Distinction from standard resampling A natural question is whether the inverse step is simply a standard remedy for posterior degeneracy (Sec. IV A). We compared it, on the same substrate (BIB-BIB, rs, Nh = 10,m= 50,T= 2×10 5), with sequential im- portance resampling (SIR), th...

  6. [14]

    V C is collected here (Fig

    Robustness-sweep figures The observation-rule ablation of Sec. V C is collected here (Fig. 13), and the window-size and game-dimension sweep of Sec. V A is shown in Fig. 12

  7. [15]

    Ezaki, E

    T. Ezaki, E. Fonseca dos Reis, T. Watanabe, M. Sakaki, and N. Masuda, Communications Biology3, 52 (2020)

    2020

  8. [16]

    P. M. M¨ ulleret al., Proceedings of the National Academy of Sciences122, e2417117122 (2025)

    2025

  9. [17]

    Y. Xin, Y. Cui, S. Yu, and N. Liu, Proceedings of the National Academy of Sciences122, e2417010122 (2025)

    2025

  10. [18]

    J. M. Beggs and D. Plenz, Journal of Neuroscience23, 11167 (2003)

    2003

  11. [19]

    Mora and W

    T. Mora and W. Bialek, Journal of Statistical Physics 144, 268 (2011)

    2011

  12. [20]

    M. A. Mu˜ noz, Reviews of Modern Physics90, 031001 (2018)

    2018

  13. [21]

    Tkaˇ cik, T

    G. Tkaˇ cik, T. Mora, O. Marre, D. Amodei, S. E. Palmer, M. J. Berry, and W. Bialek, Proceedings of the National Academy of Sciences112, 11508 (2015)

    2015

  14. [22]

    Mastromatteo and M

    I. Mastromatteo and M. Marsili, Journal of Statisti- cal Mechanics: Theory and Experiment2011, P10012 (2011)

    2011

  15. [23]

    D. J. Schwab, I. Nemenman, and P. Mehta, Physical Re- view Letters113, 068102 (2014)

    2014

  16. [24]

    Friston, M

    K. Friston, M. Breakspear, and G. Deco, Frontiers in Computational Neuroscience6, 44 (2012)

    2012

  17. [25]

    P. Bak, C. Tang, and K. Wiesenfeld, Physical Review Letters59, 381 (1987)

    1987

  18. [26]

    Bak,How Nature Works: The Science of Self- Organized Criticality(Copernicus / Springer-Verlag, 1996)

    P. Bak,How Nature Works: The Science of Self- Organized Criticality(Copernicus / Springer-Verlag, 1996)

    1996

  19. [27]

    Paczuski, S

    M. Paczuski, S. Maslov, and P. Bak, Physical Review E 53, 414 (1996)

    1996

  20. [28]

    Dickman, M

    R. Dickman, M. A. Mu˜ noz, A. Vespignani, and S. Zap- peri, Brazilian Journal of Physics30, 27 (2000)

    2000

  21. [29]

    Pruessner,Self-Organised Criticality: Theory, Mod- els and Characterisation(Cambridge University Press, 2012)

    G. Pruessner,Self-Organised Criticality: Theory, Mod- els and Characterisation(Cambridge University Press, 2012)

    2012

  22. [30]

    N. W. Watkins, G. Pruessner, S. C. Chapman, N. B. Crosby, and H. J. Jensen, Space Science Reviews198, 3 (2016)

    2016

  23. [31]

    Markovi´ c and C

    D. Markovi´ c and C. Gros, Physics Reports536, 41 (2014)

    2014

  24. [32]

    Viswanathan, S

    G. Viswanathan, S. Buldyrev, S. Havlin, M. da Luz, E. Raposo, and H. Stanley, Nature401, 911 (1999)

    1999

  25. [33]

    M. S. Abe, Proceedings of the National Academy of Sci- ences117, 24336 (2020)

    2020

  26. [34]

    Viswanathan, V

    G. Viswanathan, V. Afanasyev, S. Buldyrev, E. Murphy, P. Prince, and H. Stanley, Nature381, 413 (1996)

    1996

  27. [35]

    D. W. Simset al., Nature451, 1098 (2008)

    2008

  28. [36]

    N. E. Humphrieset al., Nature465, 1066 (2010). 25 rs ra ss sa 0.0 0.5 1.0 1.5 2.0 2.5 3.0 α (T_argmax1, best fit) (a) eq3 caseA hybrid defeatGT rs ra ss sa α (T_argmax1, best fit) (b) eq3 caseA hybrid defeatGT FIG. 13. Observation-rule ablation at medium scale (T= 10 4 ×200 ru...

    2010

  29. [37]

    I. Rhee, M. Shin, S. Hong, K. Lee, S. J. Kim, and S. Chong, IEEE/ACM Trans. on Networking19, 630 (2011)

    2011

  30. [38]

    V. M. K. Namboodiri, J. M. Levy, S. Mihalas, D. W. Sims, and M. G. Hussain Shuler, PNAS113, 8747 (2016)

    2016

  31. [39]

    Gunji, S

    Y.-P. Gunji, S. Shinohara, T. Haruna, and V. Basios, BioSystems152, 44 (2017)

    2017

  32. [40]

    Gunji, H

    Y.-P. Gunji, H. Murakami, T. Tomaru, and V. Basios, Philosophical Transactions of the Royal Society A376, 20170370 (2018)

    2018

  33. [41]

    Gunji, T

    Y.-P. Gunji, T. Kawai, H. Murakami, T. Tomaru, M. Mi- noura, and S. Shinohara, Computational and Structural Biotechnology Journal19, 247 (2021)

    2021

  34. [42]

    Gunji, BioSystems259, 105677 (2026)

    Y.-P. Gunji, BioSystems259, 105677 (2026)

    2026

  35. [43]

    Ibuka and K

    A. Ibuka and K. Sasai, inProc. 2024 IEEE Internat. Conf. on Agents (ICA)(2024) pp. 108–109

    2024

  36. [44]

    Shinohara, N

    S. Shinohara, N. Manome, K. Suzuki, U.-i. Chung, T. Takahashi, Y.-P. Gunji, Y. Nakajima, and S. Mit- suyoshi, Biosystems190, 104104 (2020)

    2020

  37. [45]

    Shinohara, N

    S. Shinohara, N. Manome, K. Suzuki, U.-i. Chung, T. Takahashi, H. Okamoto, Y.-P. Gunji, Y. Nakajima, and S. Mitsuyoshi, PLoS ONE15, e0233559 (2020)

    2020

  38. [46]

    Shinohara, N

    S. Shinohara, N. Manome, Y. Nakajima, Y.-P. Gunji, T. Moriyama, H. Okamoto, S. Mitsuyoshi, and U.-i. Chung, Symmetry13, 718 (2021)

    2021

  39. [47]

    Shinohara, H

    S. Shinohara, H. Okamoto, N. Manome, Y.-P. Gunji, Y. Nakajima, T. Moriyama, and U.-i. Chung, Chaos, Soli- tons & Fractals157, 111976 (2022)

    2022

  40. [48]

    Zinkevich, M

    M. Zinkevich, M. Johanson, M. Bowling, and C. Piccione, inAdvances in Neural Information Processing Systems, Vol. 20 (2007) pp. 1729–1736

    2007

  41. [49]

    Tammelin, arXiv preprint arXiv:1407.5042 (2014)

    O. Tammelin, arXiv preprint arXiv:1407.5042 (2014)

    2014 arXiv

  42. [50]

    Brown and T

    N. Brown and T. Sandholm, Science365, 885 (2019)

    2019

  43. [51]

    Lanctot, K

    M. Lanctot, K. Waugh, M. Zinkevich, and M. Bowling, inAdvances in Neural Information Processing Systems, Vol. 22 (2009)

    2009

  44. [52]

    Szolnoki, M

    A. Szolnoki, M. Mobilia, L.-L. Jiang, B. Szczesny, A. M. Rucklidge, and M. Perc, Journal of the Royal Society Interface11, 20140735 (2014)

    2014

  45. [53]

    T. W. Neller and M. Lanctot, An introduction to coun- terfactual regret minimization (2013), online tutorial

    2013

  46. [54]

    Zheng, J

    K. Zheng, J. Zhou, and H. Wang, arXiv preprint arXiv:2506.09390 10.48550/arXiv.2506.09390 (2025)

    2025 doi

  47. [55]

    Z. Wang, B. Xu, and H.-J. Zhou, Scientific Reports4, 5830 (2014)

    2014

  48. [56]

    K. Arai, S. Jacob, A. S. Widge, and A. Yousefi, Scientific Reports15, 14955 (2025)

    2025

  49. [57]

    G. Dong, X. Lin, H. Zhou, and X. Du, Behavioral and Brain Functions10, 11 (2014)

    2014

  50. [58]

    Xu and N

    J. Xu and N. Harvey, Cognition131, 173 (2014)

    2014

  51. [59]

    Spanknebel and K

    M. Spanknebel and K. Pawelzik, Dynamics of human co- operation in economic games (2015), arXiv preprint; no journal version published as of 2025, arXiv:1508.05288 [physics.soc-ph]

    2015 arXiv

  52. [60]

    Barab´ asi, Nature435, 207 (2005)

    A.-L. Barab´ asi, Nature435, 207 (2005)

    2005

  53. [61]

    V´ azquez, Physical Review Letters95, 248701 (2005)

    A. V´ azquez, Physical Review Letters95, 248701 (2005)

    2005

  54. [62]

    Karsai, K

    M. Karsai, K. Kaski, A.-L. Barab´ asi, and J. Kert´ esz, Sci- entific Reports2, 397 (2012)

    2012

  55. [63]

    Karsai, H.-H

    M. Karsai, H.-H. Jo, and K. Kaski,Bursty Human Dy- namics, SpringerBriefs in Complexity (Springer, 2018)

    2018

  56. [64]

    Kass, Rock Paper Scissors Lizard Spock,http: //www.samkass.com/theories/RPSSL.html(n.d.), ac- cessed: 2026-05-17

    S. Kass, Rock Paper Scissors Lizard Spock,http: //www.samkass.com/theories/RPSSL.html(n.d.), ac- cessed: 2026-05-17

    2026

  57. [65]

    Jelinek and R

    F. Jelinek and R. L. Mercer, inPattern Recognition in Practice, edited by E. S. Gelsema and L. N. Kanal (North-Holland, 1980) pp. 381–397

    1980

  58. [66]

    S. F. Chen and J. Goodman, inProc. 34th Annual Meeting of the Association for Computational Linguis- tics(1996) pp. 310–318

    1996

  59. [67]

    C. D. Manning and H. Sch¨ utze,Foundations of Statistical Natural Language Processing(MIT Press, 1999). 26

    1999

  60. [68]

    Laplace,Th´ eorie analytique des probabilit´ es (Courcier, Paris, 1812)

    P.-S. Laplace,Th´ eorie analytique des probabilit´ es (Courcier, Paris, 1812)

  61. [69]

    Alstott, E

    J. Alstott, E. Bullmore, and D. Plenz, PLoS ONE9, e85777 (2014)

    2014

  62. [70]

    Clauset, C

    A. Clauset, C. R. Shalizi, and M. Newman, SIAM Review 51, 661 (2009)

    2009

  63. [71]

    M. P. H. Stumpf and M. A. Porter, Science335, 665 (2012)

    2012

  64. [72]

    Deluca and A

    A. Deluca and A. Corral, Acta Geophysica61, 1351 (2013)

    2013

  65. [73]

    Nowak and K

    M. Nowak and K. Sigmund, Nature364, 56 (1993)

    1993

  66. [74]

    C. J. C. H. Watkins and P. Dayan, Machine Learning8, 279 (1992)

    1992

  67. [75]

    Hart and A

    S. Hart and A. Mas-Colell, Econometrica68, 1127 (2000)

    2000

  68. [76]

    T. M. Cover and J. A. Thomas,Elements of Information Theory, 2nd ed. (Wiley-Interscience, 2006)

    2006

  69. [77]

    Platt, E

    N. Platt, E. Spiegel, and C. Tresser, Physical Review Letters70, 279 (1993)

    1993

  70. [78]

    Heagy, N

    J. Heagy, N. Platt, and S. Hammel, Physical Review E 49, 1140 (1994)

    1994

  71. [79]

    Pikovsky, M

    A. Pikovsky, M. Rosenblum, and J. Kurths,Synchroniza- tion: A Universal Concept in Nonlinear Sciences(Cam- bridge University Press, 2001)

    2001

  72. [80]

    Hammer, N

    P. Hammer, N. Platt, S. Hammel, J. Heagy, and B. Lee, Physical Review Letters73, 1095 (1994)

    1994

  73. [81]

    Nakao, Physical Review E58, 1591 (1998)

    H. Nakao, Physical Review E58, 1591 (1998)

    1998

  74. [82]

    S. C. Venkataramani, T. M. Antonsen Jr., E. Ott, and J. C. Sommerer, Physica D96, 66 (1996)

    1996

  75. [83]

    Bertin, Phys

    E. Bertin, Phys. Rev. E85, 042104 (2012)

    2012

  76. [84]

    Privman and M

    V. Privman and M. Fisher, Journal of Statistical Physics 33, 385 (1983)

    1983

  77. [85]

    Gunji, Complex Systems23, 55 (2014)

    Y.-P. Gunji, Complex Systems23, 55 (2014)

    2014

  78. [86]

    Agrawal and N

    S. Agrawal and N. Goyal, inConf. on Learning Theory (COLT)(2012) arXiv:1111.1797

    2012 arXiv

  79. [87]

    Russo and B

    D. Russo and B. Van Roy, Mathematics of Operations Research39, 1221 (2014)

    2014

  80. [88]

    O’Donoghue and T

    B. O’Donoghue and T. Lattimore, inAdvances in Neu- ral Information Processing Systems, Vol. 34 (2021) pp. 12507–12519

    2021

  81. [89]

    Brockbank and E

    E. Brockbank and E. Vul, Cognitive Psychology151, 101654 (2024)

    2024

  82. [90]

    Internal- state criticality in Bayesian–inverse-Bayesian inference

    K. Sasai and Y.-P. Gunji, Simulation data for “Internal- state criticality in Bayesian–inverse-Bayesian inference”, Version 3, [Dataset], Zenodo (2026)

    2026

  83. [91]

    J. G. Oliveira and A. V´ azquez, Physica A388, 187 (2009)

    2009

  84. [92]

    N. J. Gordon, D. J. Salmond, and A. F. M. Smith, IEE Proc. F (Radar and Signal Processing)140, 107 (1993)

    1993

Pith tools

Reviewed June 26, 2026 · model on record in the stance chip above.