Bayesian Evolutionary Swarm Architecture: A Formal Epistemic System Grounded in Truth-Based Competition

Craig Steven Wright

REVIEW 5 major objections 4 minor 59 references

Bayesian Evolutionary Swarm Architecture: A Formal Epistemic System Grounded in Truth-Based Competition

T0 review · 5 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that competition among Bayesian agents under a fixed truth oracle makes truth an asymptotic evolutionary attractor.

desk verdict The central 'truth as attractor' theorem is an axiom in disguise, but the paper is a competent formal spec that could serve as a design checklist. read the letter →

arxiv 2506.19191 v1 pith:6RVCBOCU submitted 2025-06-23 cs.AI cs.CLcs.GTmath.LO

Craig Steven Wright This is my paper · ORCID

classification cs.AIcs.CLcs.GTmath.LO MSC 68T0560J2062F1591A22

keywords Bayesianinferenceswarmintelligenceevolutionarydynamicstruthoraclepopulationconvergencetheoremagentratingsystemproperscoringrules

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

This paper tries to show that a swarm of Bayesian agents, each holding a probability distribution over hypotheses, can be wired so that truth wins in the long run. Agents are scored by an external truth oracle, their ratings rise and fall with how well their beliefs align with that oracle, and high-rated agents reproduce while low-rated agents are eventually eliminated. The central result, Theorem 20.1, asserts that if the oracle functional has a unique best hypothesis and agent utility is a monotone transformation of oracle score plus bounded noise, then the population-weighted belief mass concentrates around that hypothesis in total variation as time goes to infinity. If the theorem is right, verifiable knowledge emerges from competitive pressure rather than from a centralised designer.

What carries the argument

The load-bearing object is the coupled Markovian population process over belief measures and ratings, built from selection, reproduction with rating attenuation, prior mutation, delayed extinction, and Bayesian posterior updates. The identity that makes the argument work is Axiom 20.1, $u_i(h,t)=f(O(h))+\epsilon_{i,t}$: agent utility is a monotone transformation of oracle score plus bounded noise. That decomposition turns the oracle into a fitness gradient, and the convergence then runs through a multiplicative-weight mechanism in which agents with high oracle scores leave more descendants carrying perturbed copies of their beliefs.

What would settle it

Simulate the process under Theorem 20.1's assumptions with a two-point hypothesis space $H=\{h^*, h'\}$, oracle scores $O(h^*)=1$ and $O(h')=0.99$, and all agents initially placing most mass on $h'$. If the total-variation distance between the population-weighted belief and the point mass at $h^*$ does not go to zero as $t$ grows, the convergence claim is false. If it does, the attractor claim passes this concrete test.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is Theorem 20.1, stated in Section 12.3. It says: assume agents reproduce with probability increasing in rating, beliefs evolve through stochastic updates directed by reward gradients, and the oracle functional $O$ has a unique global maximiser $h^*$ in $H$. Then, in the limit $t \to \infty$, the population-weighted belief mass satisfies $\sum_i \frac{R_i(t)}{\sum_j R_j(t)} \mu_i(t)(B_\varepsilon(h^*)) \to 1$ for every $\varepsilon > 0$. In words, the population's belief, weighted by ratings, converges in total variation to a point mass at the oracle's best hypothesis. The paper interprets this as truth being an evolutionary attractor: agents near $h^*$ out-reproduce others, so the dynamics of selection and inheritance do the epistemic work.

Load-bearing premise

The theorem only applies if agent utility really is a monotone transformation of oracle score plus bounded noise (Axiom 20.1); if a real task does not supply such oracle-aligned reward, the convergence result has nothing to act on.

Editorial extensions

If this is right

If Theorem 20.1 is correct, truth-aligned agents become evolutionarily dominant and adversarial or noise-aligned configurations are asymptotically suppressed.
The system asymptotically penalises epistemic divergence from $h^*$: internally consistent but externally ungrounded beliefs do not survive in the limit.
Under the paper's bounded-perturbation and reproductive-stability hypotheses, ratings converge in distribution to a non-degenerate limit and expected population size remains bounded.
With contraction-based local update operators and bounded communication intervals, asynchronous updates converge to unique fixed points for each agent.
The attenuation condition $\lambda < 1/2$ guarantees that the agent count remains finite under reproduction.

Reading between the lines

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

Editorial inference: the same mechanism predicts that if the reward signal tracks something other than the stated oracle, say a biased metric, the swarm will converge to that reward's maximiser instead, so the architecture transfers to whatever objective the oracle encodes.
Editorial inference: the framework suggests a testable recipe for model ensembling: maintain a population of Bayesian models, score them with a proper scoring rule, and let rating-based reproduction replace explicit gradient-based training.
Editorial inference: Theorem 20.1 is asymptotic, and nothing in the paper bounds the time to concentration, so a natural next step is to derive finite-time concentration rates in terms of noise variance, mutation scale, and population size.
Editorial inference: the multiplicative-weight flavour of the convergence argument invites a connection to regret bounds for exponential weights, which could give the theorem a non-asymptotic companion.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

I've read the paper and your notes. The short version: the headline claim—truth as an evolutionary attractor—is effectively an axiom in disguise, and the one theorem that matters (20.1) is a proof sketch that doesn't do the work. But there's more here than a crank submission. It's a careful, extensive specification of an evolutionary swarm, and if you read it as a design checklist rather than a proof system, some parts are useful.

What's actually good: the paper builds a coherent architecture from standard pieces—Bayesian posteriors, replicator dynamics, rating thresholds, hash commitments—and states them in a uniform measure-theoretic language. The definitions of the oracle, selection/reproduction/extinction operators, and the security layer are mutually consistent. Several lemmas are correctly lifted from standard references. As a formal exercise in 'here is how you would write down such a system,' it's competent.

The soft spots are where the load sits. Theorem 8.1 claims an iff between lambda < 1/2 and finite population, but the proof tracks rating mass; mass decay doesn't bound agent count. That inference is simply false. Theorem 20.1 is the centerpiece, and it's a three-sentence sketch. Worse, Axiom 20.1 defines utility as f(O(h)) plus noise, so the conclusion that mass concentrates near argmax O is already baked into the fitness function. That's circular. The stress-test note adds a fair point: even granting the axiom, the stated dynamics don't preclude a mutation operator with invariant subsets away from h*, or thresholds that stop reproduction entirely. So the theorem is unproven within the model. Several other theorems are labeled sketches or just cite literature.

There's no implementation or data, and the free parameters (thresholds, noise, learning rates) are uncalibrated, so the operational claims are unsupported.

Net: I'd treat this as a reference design spec, not a research result. It could be background for someone building an auditable multi-agent learning system. But as a paper claiming formal guarantees, it doesn't hold up. My recommendation: desk reject, with an invitation to resubmit if the central theorem is actually proved—which would require adding real dynamics and removing the circularity. The author is a serious thinker, but this submission isn't there yet.

Referee Report

5 major / 4 minor

Summary. The paper proposes a formal architecture for a swarm-based AI in which Bayesian agents compete under evaluation by an external truth oracle, with ratings driving reproduction and extinction. It introduces measure-theoretic belief spaces, Markovian rating dynamics, mutation operators, cryptographic identity commitments, and a series of theorems claiming convergence, stability, and security. The central claim is Theorem 20.1 in §12.3, which asserts that under oracle-aligned utilities and a unique oracle maximizer h*, the population-weighted belief mass concentrates around h* in total variation as t→∞.

Significance. If substantiated, the claim that truth is an asymptotic evolutionary attractor would be a significant formal result linking Bayesian inference, evolutionary dynamics, and epistemic utility. The paper also provides useful formal scaffolding: explicit measurability and computability axioms, regular conditional probability foundations, bounded rating processes, and cryptographic hashing of agent state. However, the central theorem is not proven; its proof is a short sketch, and the main mechanism is assumed through Axiom 20.1 rather than derived. The manuscript therefore does not currently deliver a rigorous guarantee of truth convergence, only a formal language in which such a claim could be stated.

major comments (5)

[§12.3, Theorem 20.1] The central theorem is asserted with a three-sentence proof sketch. It does not follow from assumptions (i)–(iii) and Axiom 20.1 that population-weighted belief mass converges in total variation to the oracle maximizer h*. No state space for the claimed Markov chain is defined, no invariant measure or Lyapunov function is exhibited, and no argument rules out dynamics that fail to concentrate. For instance, the mutation operator of §6.2 can be chosen with a closed invariant subset not containing h*; the reproduction threshold θ_spawn can be set so high that no agent reproduces; and the middle-agent equilibrium of §5.3 is a stationary rating distribution that need not coincide with the oracle maximizer. All of these configurations are compatible with Axiom 20.1 and with assumptions (i)–(iii), so the theorem as stated is false or at least not established.
[§6.1, Theorem 8.1] The Spawning Stability Theorem claims that agent count remains finite if and only if λ < 1/2, but the proof only demonstrates a necessary condition on rating mass. The equations N_{t+1} = 2N_t and M_{t+1} = 2λM_t are compatible: rating mass can decay to zero while the agent count doubles without bound. Thus the claimed 'if and only if' does not follow from the given argument, and the theorem does not establish a finite bound on the number of agents.
[§4.2 and §12.3, Axiom 20.1] Axiom 20.1 assumes agent utility is a monotone function of the oracle score plus bounded noise, u_i(h,t) = f(O(h)) + ε_{i,t}. This places the desired conclusion into the assumptions: if fitness is already defined to increase with the oracle score, then selection-for-truth is true by construction. The theorem therefore does not provide an independent bridge from competition to truth alignment. The manuscript also does not specify how such an oracle-aligned utility could be obtained for real tasks; without that, the convergence result has no demonstrated domain of application beyond the axiom itself.
[§7.2 and §8.4, Theorems 10.1 and 12.3] Several formal results are supported only by citations or proof sketches rather than complete arguments. Theorem 10.1 on almost sure convergence to epistemic attractors invokes stochastic approximation theory without verifying its conditions, and Theorem 12.3 on bifurcation-induced phase shifts is justified by a sketch plus classical bifurcation citations. Since these theorems are presented as part of a 'mathematically rigorous' framework, the proofs need to be supplied or the results should be explicitly labelled as conjectures.
[§7.3, Corollary 10.2.1] The corollary claims that the stationary rating distribution concentrates around a truth-aligned band [1−ε,1], but its proof sketch is one sentence and does not follow from the preceding quasi-stationary convergence theorem, which only establishes weak convergence to some ν* satisfying a martingale-type condition. Concentration on high ratings requires additional assumptions about extinction, mutation, and the utility gradients that are not stated or proved.

minor comments (4)

[General notation] The paper uses inconsistent names for the same parameters, e.g., θ_spawn and τ_rep, θ_death and τ_ext, λ and β for attenuation, which makes verification of the formal statements unnecessarily difficult.
[§2.4 and §12.3] The truth oracle is defined in §2.4 as a two-argument dissimilarity function T(ŷ, y), but in §12.3 the oracle is a single-argument functional O:H→R. These two formalizations are never reconciled, leaving the relationship between oracle loss ℓ_i(t) and oracle score O(h) implicit.
[§12.1] The summary of results refers to theorems and axioms by numbers that do not match the numbering used earlier in the text (e.g., 'Theorem 9.2' and 'Definition 15.2'), which will confuse readers attempting to verify the claims.
[Throughout] Several definitions are labelled 'Axiom' when they are merely design choices rather than foundational assumptions, e.g., Axiom 54 (Population Preservation Axiom) and Axiom 63 (Entropy Regularisation Principle). This inflates the logical status of the model and should be revised.

Circularity Check

1 steps flagged · score 8.0 of 10

Central 'truth as evolutionary attractor' theorem restates the truth-aligned fitness axiom: utility is defined as a monotone function of oracle score, so selection toward the oracle maximizer is built in by construction.

self definitional [Section 12.3, Axiom 20.1 and Theorem 20.1; Section 4.4, Axiom 47]
"Axiom 20.1: "Then there exists a monotonic transformation f:R→R such that: u_i(h,t)=f(O(h))+ε_i,t, where ε_i,t∼N(0,σ^2) models bounded observational noise." Axiom 47: "P(replicate|a_i,t)∝max{0,∇R_i(t)}." Theorem 20.1: "Then in the limit t→∞, the population-weighted belief mass concentrates around h* in total variation.""

Axiom 20.1 makes utility a monotone function of the oracle score O(h) plus noise, and Axiom 47 makes reproduction probability proportional to the reward gradient derived from that utility. Therefore the type with maximal O(h) is, by definition, the type with maximal reproductive propensity. Theorem 20.1's conclusion that population-weighted belief mass concentrates on the oracle maximizer h* is the fitness definition restated as a dynamic claim. The proof sketch ('Markov chain with absorbing tendencies near h*', 'multiplicative weight update mechanism') supplies no mechanism beyond the already-assumed truth-aligned selection gradient. The 'attractor' is installed in the fitness function before any dynamics are analysed.

full rationale

The paper's central claim—'the system establishes truth as an evolutionary attractor'—reduces to its own definition of fitness. Axiom 20.1 defines agent utility as a monotone transformation of the oracle score O(h) plus noise; Section 4.4 (Axiom 47) makes reproduction probability proportional to max{0,∇R_i(t)}, where ∇R_i(t) is the reward gradient built from truth-aligned utility. Thus the oracle maximizer h* is, by construction, the type with maximal reproductive propensity. Theorem 20.1's conclusion that population-weighted belief mass concentrates around h* is an unpacking of that construction. No independent empirical input or external benchmark is involved; if O were replaced by any arbitrary functional G, the same argument would 'prove' that the swarm converges to argmax G. The paper's only actual evidence for the theorem is a three-sentence proof sketch invoking 'absorbing tendencies near h*' and an undefined 'multiplicative weight update mechanism'; this is an omitted proof, which we do not score as circularity but which removes any independent mathematical content beyond the definitional loading. There are no load-bearing self-citations: the references are standard textbooks and papers. The same definitional pattern appears in Theorem 10.1, where a 'truth-aligned reward landscape' is assumed and convergence to 'truth-aligned invariant measures maximising expected utility' is concluded. Score 8: the central result is forced by definition, not derived.

Assumptions & free parameters 6 free parameters · 6 assumptions · 1 invented entities

The system's guarantees depend on hand-chosen thresholds and coefficients, none of which are estimated from data. The most important free choice is the truth oracle itself, which is assumed rather than constructed.

free parameters (6)

reproduction threshold theta_spawn
Chosen in (0,1) to define eligibility for spawning; central to selection dynamics (Definition 4.2).
extinction threshold theta_death
Chosen in (0,1) to define when agents are marked for extinction; couples with grace period (Definition 4.4).
attenuation coefficient lambda
Controls offspring rating decay; Theorem 8.1 claims population finiteness iff lambda < 1/2, but the proof is internally inconsistent.
noise variance sigma^2
Added to rating updates and utility; convergence theorems rely on bounded noise but no value or estimator is given.
entropy regularization weight beta
Chosen in R+ to add diversity pressure; Theorem 10.3 depends on it without a calibration procedure.
learning rate alpha_t = non-increasing, e.g., 1/(t+1)
Rating update uses alpha_t; convergence rates depend on it but it is left unspecified in most theorems.

assumptions (6)

domain assumption Existence of an external, immutable truth oracle O with a unique global maximizer h* (Axioms 22, 24; Theorem 20.1 assumption iii)
The central convergence result presupposes a fixed ground-truth evaluator that agents can only access through indirect feedback; no construction or existence proof is given.
ad hoc to paper Truth fitness monotonicity: u_i(h,t) = f(O(h)) + epsilon_i,t for a monotone f (Axiom 20.1)
This axiom already equates agent utility with oracle alignment, so the truth-as-attractor theorem is a consequence of the definition, not a derived result.
ad hoc to paper Entropy regularization principle with fixed beta (Axiom 63)
Added to guarantee diversity and prevent collapse; its specific form is chosen for the theorem and not justified by data.
ad hoc to paper Limit Existence Axiom: tightness and asymptotic time-homogeneity imply a unique limit distribution (Axiom 62)
The existence of the limit distribution is assumed under compactness conditions rather than proven from the dynamics.
ad hoc to paper Population preservation: total rating mass non-increasing under reproduction (Axiom 54)
Used in Theorem 8.1; the proof then contradicts itself by allowing unbounded doubling under the same conditions.
standard math ZFC and Polish-space measurability (Axioms 1-6)
Background foundation for measure theory and Bayesian updating; not controversial.

invented entities (1)

Truth Oracle O
purpose: Exogenous functional that scores agent predictions against ground truth and drives all selection; the load-bearing epistemic standard.
No construction, benchmark, or falsifiable prediction is provided; the system's guarantees hold only if such an oracle exists and is immutable.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Bayesian Evolutionary Swarm Architecture: A Formal Epistemic System Grounded in Truth-Based Competition." pith.science (2026). https://pith.science/paper/6RVCBOCU

@misc{pith2026250619191,
  author       = {Pith},
  title        = {Pith review of: Bayesian Evolutionary Swarm Architecture: A Formal Epistemic System Grounded in Truth-Based Competition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6RVCBOCU}},
  note         = {Machine review of arXiv:2506.19191}
}

read the original abstract

We introduce a mathematically rigorous framework for an artificial intelligence system composed of probabilistic agents evolving through structured competition and belief revision. The architecture, grounded in Bayesian inference, measure theory, and population dynamics, defines agent fitness as a function of alignment with a fixed external oracle representing ground truth. Agents compete in a discrete-time environment, adjusting posterior beliefs through observed outcomes, with higher-rated agents reproducing and lower-rated agents undergoing extinction. Ratings are updated via pairwise truth-aligned utility comparisons, and belief updates preserve measurable consistency and stochastic convergence. We introduce hash-based cryptographic identity commitments to ensure traceability, alongside causal inference operators using do-calculus. Formal theorems on convergence, robustness, and evolutionary stability are provided. The system establishes truth as an evolutionary attractor, demonstrating that verifiable knowledge arises from adversarial epistemic pressure within a computable, self-regulating swarm.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 54 canonical work pages

[1]

Ash and Catherine A

Robert B. Ash and Catherine A. Doléans-Dade.Probability and Measure Theory. Academic Press, 2000

work page 2000
[2]

Athreya and Peter E

Krishna B. Athreya and Peter E. Ney.Branching Processes. Springer-Verlag, 1972

work page 1972
[3]

Random oracles are practical: A paradigm for designing efficient protocols

Mihir Bellare and Phillip Rogaway. Random oracles are practical: A paradigm for designing efficient protocols. InProceedings of the 1st ACM Conference on Computer and Communications Security, pages 62–73. ACM, 1993

work page 1993
[4]

Dynamics of stochastic approximation algorithms.Séminaire de Probabilités XXXIII, 1709:1–68, 1999

Michel Benaïm. Dynamics of stochastic approximation algorithms.Séminaire de Probabilités XXXIII, 1709:1–68, 1999

work page 1999
[5]

Bernardo

José M. Bernardo. Expected information as expected utility.Annals of Statistics, 7(3):686–690, 1979

work page 1979
[6]

Bernardo and Adrian F

José M. Bernardo and Adrian F. M. Smith.Bayesian Theory. Wiley, 2000

work page 2000
[7]

Wiley-Interscience, 3 edition, 1995

Patrick Billingsley.Probability and Measure. Wiley-Interscience, 3 edition, 1995

work page 1995
[8]

John Wiley & Sons, 2 edition, 1999

Patrick Billingsley.Convergence of Probability Measures. John Wiley & Sons, 2 edition, 1999

work page 1999

Show all 59 references

[9]

Bogachev.Measure Theory, volume I–II

Vladimir I. Bogachev.Measure Theory, volume I–II. Springer, 2007

2007
[10]

Springer, 1986

Lucien Le Cam.Asymptotic Methods in Statistical Decision Theory. Springer, 1986

1986
[11]

Springer, 2018

René Carmona and François Delarue.Probabilistic Theory of Mean Field Games with Applications I: Mean Field FBSDEs, Control, and Games, volume 83 ofProbability Theory and Stochastic Modelling. Springer, 2018

2018
[12]

Updating probabilities.AIP Conference Proceedings, 872(1):31–42, 2006

Ariel Caticha and Adom Giffin. Updating probabilities.AIP Conference Proceedings, 872(1):31–42, 2006

2006
[13]

Cam- bridge University Press, 2006

Nicolo Cesa-Bianchi and Gabor Lugosi.Prediction, Learning, and Games. Cam- bridge University Press, 2006

2006
[14]

Chee-Yee Chong and S. P. Kumar. Distributed tracking in sensor networks.Pro- ceedings of the IEEE, 91(8):1241–1257, 2003

2003
[15]

Cover and Joy A

Thomas M. Cover and Joy A. Thomas.Elements of Information Theory. Wiley, 1991

1991
[16]

Cover and Joy A

Thomas M. Cover and Joy A. Thomas.Elements of Information Theory. Wiley, 2nd edition, 2006. 79

2006
[17]

I-divergence geometry of probability distributions and minimization problems.The Annals of Probability, 3(1):146–158, 1975

Imre Csiszár. I-divergence geometry of probability distributions and minimization problems.The Annals of Probability, 3(1):146–158, 1975

1975
[18]

Finitely additive probabilities.Annales de l’Institut Henri Poincaré, 7:1–68, 1937

Bruno de Finetti. Finitely additive probabilities.Annales de l’Institut Henri Poincaré, 7:1–68, 1937

1937
[19]

R. M. Dudley.Real Analysis and Probability, volume 74 ofCambridge Studies in Advanced Mathematics. Cambridge University Press, 2 edition, 2002

2002
[20]

Cambridge University Press, 4 edition, 2010

Rick Durrett.Probability: Theory and Examples. Cambridge University Press, 4 edition, 2010

2010
[21]

Computable solutions of domain equations

Abbas Edalat and Philipp Sünderhauf. Computable solutions of domain equations. Theoretical Computer Science, 210(1):73–98, 1999

1999
[22]

Enderton.Elements of Set Theory

Herbert B. Enderton.Elements of Set Theory. Academic Press, 1977

1977
[23]

Ethier and Thomas G

Stewart N. Ethier and Thomas G. Kurtz.Markov Processes: Characterization and Convergence. John Wiley & Sons, 1986

1986
[24]

Ferguson

Thomas S. Ferguson. A bayesian analysis of some nonparametric problems.Annals of Statistics, 1(2):209–230, 1973

1973
[25]

The free-energy principle: a unified brain theory?Nature Reviews Neuroscience, 11(2):127–138, 2010

Karl Friston. The free-energy principle: a unified brain theory?Nature Reviews Neuroscience, 11(2):127–138, 2010

2010
[26]

Pac- bayesian theory meets bayesian inference.Advances in Neural Information Process- ing Systems, 29:1884–1892, 2016

Pascal Germain, Amaury Habrard, François Laviolette, and Emilie Morvant. Pac- bayesian theory meets bayesian inference.Advances in Neural Information Process- ing Systems, 29:1884–1892, 2016

2016
[27]

Cambridge University Press, 2017

Subhashis Ghosal and Aad van der Vaart.Fundamentals of Nonparametric Bayesian Inference. Cambridge University Press, 2017

2017
[28]

Tilmann Gneiting and Adrian E. Raftery. Strictly proper scoring rules, prediction, and estimation.Journal of the American Statistical Association, 102(477):359–378, 2007

2007
[29]

Springer, 1983

John Guckenheimer and Philip Holmes.Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields. Springer, 1983

1983
[30]

Halmos.Measure Theory

Paul R. Halmos.Measure Theory. D. Van Nostrand Company, 1950

1950
[31]

David Haussler, Jyrki Kivinen, and Manfred K. Warmuth. Sequential prediction using a dynamically updated predictive mixture.Journal of Computer and System Systems, 55(1):114–129, 1998. 80

1998
[32]

Cambridge University Press, 1998

Josef Hofbauer and Karl Sigmund.Evolutionary Games and Population Dynamics. Cambridge University Press, 1998

1998
[33]

Springer, 2 edition, 2002

Olav Kallenberg.Foundations of Modern Probability. Springer, 2 edition, 2002

2002
[34]

Springer, 2 edition, 2013

Achim Klenke.Probability Theory: A Comprehensive Course. Springer, 2 edition, 2013

2013
[35]

A. N. Kolmogorov.Foundations of the Theory of Probability. Chelsea Publishing Company, 2 edition, 1956

1956
[36]

Konda and John N

Vijay R. Konda and John N. Tsitsiklis. On actor-critic algorithms.SIAM Journal on Control and Optimization, 42(4):1143–1166, 2003

2003
[37]

College Publications, 2011

Kenneth Kunen.Set Theory. College Publications, 2011

2011
[38]

Kushner and G

Harold J. Kushner and G. George Yin.Stochastic Approximation and Recursive Algorithms and Applications, volume 35 ofApplications of Mathematics. Springer, 2nd edition, 2003

2003
[39]

Mean field games.Japanese Journal of Mathematics, 2(1):229–260, 2007

Jean-Michel Lasry and Pierre-Louis Lions. Mean field games.Japanese Journal of Mathematics, 2(1):229–260, 2007

2007
[40]

Dennis V. Lindley. On a measure of the information provided by an experiment. The Annals of Mathematical Statistics, 27(4):986–1005, 1956

1956
[41]

Learning social behavior.Robotics and Autonomous Systems, 20(2- 4):191–204, 1997

Maja J Mataric. Learning social behavior.Robotics and Autonomous Systems, 20(2- 4):191–204, 1997

1997
[42]

Meyn and Richard L

Sean P. Meyn and Richard L. Tweedie.Markov Chains and Stochastic Stability. Springer-Verlag, 1993

1993
[43]

Mézard, G

M. Mézard, G. Parisi, and M. A. Virasoro.Spin Glass Theory and Beyond. World Scientific, 1987

1987
[44]

Springer, 2016

Sylvie Méléard.Quasi-Stationary Distributions: Markov Chains, Diffusions and Dynamical Systems. Springer, 2016

2016
[45]

Narendra and Mandayam A

Kumpati S. Narendra and Mandayam A. L. Thathachar.Learning Automata: An Introduction. Prentice-Hall, 1989

1989
[46]

Bayesian nonparametric models.Encyclopedia of Machine Learning, pages 81–89, 2011

Peter Orbanz and Yee Whye Teh. Bayesian nonparametric models.Encyclopedia of Machine Learning, pages 81–89, 2011

2011
[47]

World Scientific, 1990

Giorgio Parisi.Mean Field Theory of Spin Glasses: Statistical Field Theory and Renormalization. World Scientific, 1990. 81

1990
[48]

K. R. Parthasarathy.Probability Measures on Metric Spaces. Probability and Math- ematical Statistics. Academic Press, 1967

1967
[49]

Cambridge University Press, 2nd edition, 2009

Judea Pearl.Causality: Models, Reasoning and Inference. Cambridge University Press, 2nd edition, 2009

2009
[50]

Cryptographic hash-function basics: Def- initions, implications, and separations for preimage resistance, second preimage re- sistance, and collision resistance

Phillip Rogaway and Thomas Shrimpton. Cryptographic hash-function basics: Def- initions, implications, and separations for preimage resistance, second preimage re- sistance, and collision resistance. InFast Software Encryption (FSE), volume 3017 ofLecture Notes in Computer Sci...

2004
[51]

Yanofsky, and Noah D

Daniel Roy, Charles E. Yanofsky, and Noah D. Goodman. Computability, inference and modelling in probabilistic programming.arXiv preprint arXiv:1206.3255, 2008

2008 arXiv
[52]

Leonard J. Savage. Elicitation of personal probabilities and expectations.Journal of the American Statistical Association, 66(336):783–801, 1971

1971
[53]

Cengage Learning, 2012

Michael Sipser.Introduction to the Theory of Computation. Cengage Learning, 2012

2012
[54]

Taylor and Leo B

Peter D. Taylor and Leo B. Jonker. Evolutionary stable strategies and game dynam- ics.Mathematical Biosciences, 40(1–2):145–156, 1978

1978
[55]

Springer, 2008

Cédric Villani.Optimal Transport: Old and New, volume 338 ofGrundlehren der mathematischen Wissenschaften. Springer, 2008

2008
[56]

Wainwright and Michael I

Martin J. Wainwright and Michael I. Jordan.Graphical Models, Exponential Fami- lies, and Variational Inference. Foundations and Trends in Machine Learning. Now Publishers Inc., 2008

2008
[57]

Weibull.Evolutionary Game Theory

Jörgen W. Weibull.Evolutionary Game Theory. MIT Press, 1995

1995
[58]

Springer, 2000

Klaus Weihrauch.Computable Analysis: An Introduction. Springer, 2000

2000
[59]

Information-theoretic analysis of generaliza- tion capability of learning algorithms.Advances in Neural Information Processing Systems, 30:2524–2533, 2017

Anastasios Xu and Maxim Raginsky. Information-theoretic analysis of generaliza- tion capability of learning algorithms.Advances in Neural Information Processing Systems, 30:2524–2533, 2017. 82

2017

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Ash and Catherine A

Robert B. Ash and Catherine A. Doléans-Dade.Probability and Measure Theory. Academic Press, 2000

work page 2000

[2] [2]

Athreya and Peter E

Krishna B. Athreya and Peter E. Ney.Branching Processes. Springer-Verlag, 1972

work page 1972

[3] [3]

Random oracles are practical: A paradigm for designing efficient protocols

Mihir Bellare and Phillip Rogaway. Random oracles are practical: A paradigm for designing efficient protocols. InProceedings of the 1st ACM Conference on Computer and Communications Security, pages 62–73. ACM, 1993

work page 1993

[4] [4]

Dynamics of stochastic approximation algorithms.Séminaire de Probabilités XXXIII, 1709:1–68, 1999

Michel Benaïm. Dynamics of stochastic approximation algorithms.Séminaire de Probabilités XXXIII, 1709:1–68, 1999

work page 1999

[5] [5]

Bernardo

José M. Bernardo. Expected information as expected utility.Annals of Statistics, 7(3):686–690, 1979

work page 1979

[6] [6]

Bernardo and Adrian F

José M. Bernardo and Adrian F. M. Smith.Bayesian Theory. Wiley, 2000

work page 2000

[7] [7]

Wiley-Interscience, 3 edition, 1995

Patrick Billingsley.Probability and Measure. Wiley-Interscience, 3 edition, 1995

work page 1995

[8] [8]

John Wiley & Sons, 2 edition, 1999

Patrick Billingsley.Convergence of Probability Measures. John Wiley & Sons, 2 edition, 1999

work page 1999