Pith. sign in

REVIEW 6 major objections 6 minor 44 references

MAGNET-KG: Maximum-Entropy Geometric Networks for Temporal Knowledge Graphs: Theoretical Foundations and Mathematical Framework

T0 review · 6 major / 6 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read This paper claims that maximum-entropy principles uniquely determine both the scoring function and the temporal likelihood for knowledge graphs, yielding a log-linear score and the complementary log-log link.

desk verdict Corollary 4.1 overreaches: the global score function does not follow from the per-head MaxEnt theorem; the paper is a useful collection of known results with a few genuine errors. read the letter →

arxiv 2509.10587 v1 pith:5PV26T43 submitted 2025-09-12 cs.IT math.IT

Ibne Farabi Shihab This is my paper · ORCID
classification cs.ITmath.IT MSC 62B1094A1762M0560G10
keywords temporalknowledgegraphsmaximumentropycomplementarylog-loglinkhyperbolicgeometrymixture-of-metricsgeneralizationboundsbeta-mixingbininvariance
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 put temporal knowledge graph link prediction on a first-principles footing. It argues that if you want the least-biased distribution over candidate tails given observed moment constraints on graph structure and geometric distance, the distribution must be log-linear, which uniquely induces a canonical score. It further argues that the only temporal Bernoulli likelihood that survives arbitrary time-bin refinement is the complementary log-log link. In sparse hierarchical regimes, it derives necessity results that push toward hyperbolic geometry, and it gives explicit generalization bounds. If true, the paper removes many free design choices from temporal knowledge graph models.

What carries the argument

Maximum entropy with moment constraints: maximize Shannon entropy over the probability simplex subject to the empirical means of the composite squared distance D and the graph support feature bS being matched. The log-sum-exp dual is strictly convex, and a full-rank feature matrix F = [1, bS_i, D_i] guarantees unique Lagrange multipliers, producing the exponential-family score. The temporal side rests on treating the survival probability as infinitely divisible: the independent-increments condition becomes Cauchy's functional equation, whose only continuous solution is exponential, and the additive-predictor form then forces the cloglog link.

What would settle it

Take a candidate set with K=3 and features (bS, d^2) = (1,1),(2,2),(3,3), as in the paper's Example 9.1; here the moment constraints are linearly dependent and the MaxEnt solution is not unique, so Corollary 4.1 does not apply. More generally, any real sparse temporal knowledge graph where the feature matrix has rank less than 3 shows the theorem's precondition failing, and any simulated non-stationary series with heavy-tailed event persistence would violate the beta-mixing condition that underlies the consistency results.

Watch

Extended reading notes

Core claim

The centerpiece is a double uniqueness claim. First, under moment constraints on a composite squared distance and a graph-structural feature, the maximum-entropy distribution over tail candidates must be exponential-family, so the score function is forced to take the form f_u(h,r,t) = alpha_{r,u} + beta_r * support(h,r,t;u) - tau_r * distance^2(h,r,t;u), with no other functional form admitted. Second, if the score is used as the log-intensity of a temporal point process, then bin invariance—where any coarsening or refinement of time bins preserves survival probabilities—forces the Bernoulli observation model to be P(Y=1|f) = 1 - exp(-Delta * e^{a f + b}), the complementary log-log link. Thes

Load-bearing premise

The central claims collapse if the feature matrix [1, bS_i, d^2_i] does not have full rank or the moment constraints are infeasible for some candidate set, and they also depend on the temporal process being strictly stationary and beta-mixing with sum of beta(k)^{1/3} finite.

Editorial extensions

If this is right

  • Any temporal link predictor consistent with these principles should use the log-linear score f = alpha + beta*support - tau*distance^2; no other functional form is allowed under the stated moment constraints.
  • The temporal observation model is forced to be the complementary log-log link rather than logistic or probit; those alternative links are excluded by bin invariance.
  • Sparse hierarchical temporal structures require a hyperbolic embedding component; Euclidean-only embeddings suffer distortion that grows without bound in the worst case.
  • Mixture weights over Euclidean, hyperbolic, and spherical geometries converge to the minimum-distortion geometry as the temperature goes to zero, making geometry selection automatic.
  • Under beta-mixing temporal dependence, the framework yields an effective sample size of at least O(N^{2/3}) and explicit generalization bounds with constants.

Reading between the lines

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

  • If the uniqueness theorems are correct, then existing temporal KG models that use logistic or softmax link functions are outside the class of bin-invariant likelihoods; a synthetic experiment comparing cloglog against logistic under Poisson binning would directly test which is better calibrated.
  • The paper's informal dimension heuristic d = Omega(log n / log(1/(1-rho))) could be made rigorous by connecting it to the covering-number bound and tested empirically through underparameterization curves.
  • Remark 12.1 concedes that the geometric-flow analogy is not rigorously derived; measuring whether gradient descent on the cloglog loss tracks distortion-energy descent would either seal or break that connection.
  • The failure-mode section of the paper itself warns that non-mixing, extreme sparsity, and non-stationarity can void the guarantees; in practice the diagnostics it lists, such as rank(F) < 3, should be checked before relying on the uniqueness and generalization results.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

6 major / 6 minor

Summary. The paper proposes a maximum-entropy (MaxEnt) foundation for temporal knowledge graph scoring, claiming a unique canonical score f_u(h,r,t)=α_{r,u}+β_r bS_u(h,r,t)-τ_r d^2(h,r,t;u), a unique bin-invariant cloglog likelihood, geometric necessity results, explicit generalization bounds, and consistency under temporal dependence. The MaxEnt derivation for a fixed candidate set and the cloglog characterization are classical and largely standard, but the paper's central step—passing from per-head MaxEnt multipliers to a global canonical score—is unsupported, and several auxiliary theorems contain incorrect statements. The advertised unified framework therefore does not hold as presented.

Significance. If correct, the framework would provide a principled unification of MaxEnt scoring, temporal likelihood choice, and geometric mixture selection for TKGs, with explicit generalization constants. The paper deserves credit for presenting a self-contained mathematical setup, for making assumptions explicit, for giving a full uniqueness proof of the cloglog link under partition invariance, and for listing failure modes and diagnostics. However, the central uniqueness claim is not proven and appears false under the stated per-head constraints; the cloglog Lipschitz bound is incorrect; and the ranking-risk theorem does not follow from the squared-distance score. These are load-bearing issues, so the paper as a whole is not acceptable in its current form.

major comments (6)
  1. [§4, Remark 4.1 and Algorithm 1] Theorem 4.2 solves the MaxEnt problem for each fixed (h,r,u) over candidate tails C_{h,r,u}. Its KKT multipliers (α*,β*,τ*) are therefore head-specific. Corollary 4.1 asserts a 'uniquely induced' global score with head-independent β_r,τ_r; nothing in the proof shows the multipliers coincide across heads. The empirical moments (5)-(6) aggregate over all observed (h,t) for (r,u), while the stationarity conditions (13)-(14) are taken over the tails of a single h. Requiring E_{p_h}[bS]=c^{(S)}_{r,u} and E_{p_h}[D]=c^{(D)}_{r,u} simultaneously for every h imposes four equations on the two global parameters (β_r,τ_r) already with two heads; Assumption 4.1, rank(F)=3, does not prevent infeasibility. Thus the central 'unique canonical score function' does not follow from Theorem 4.2 and is false in general under the stated assumptions.
  2. [§8, Theorem 8.2] The circular dependency between the MaxEnt constraints and the composite energy D is not resolved. Theorem 4.2 is for a fixed D, but D depends on embeddings, transports, and mixture weights that are subsequently learned. Algorithm 1 alternates MaxEnt solves with gradient/softmax updates, yet Theorem 3.2 only guarantees stationarity of the surrogate objective J; it does not show that any limit point satisfies the MaxEnt moment constraints (7)-(8). A fixed-point theorem for the alternating scheme is needed before one can claim that the final model is the unique MaxEnt distribution characterized in Theorem 4.2.
  3. [§11, Theorem 11.1] The stated Lipschitz constant for the cloglog loss is incorrect. For y=1, d/df[-log(1-e^{-Δe^f})] = -A e^{-A}/(1-e^{-A}) = -A/(e^A-1) with A=Δe^f, so the absolute derivative is at most 1. For y=0, the derivative is A ≤ Δ_max e^{F_max}. Hence L_ℓ ≤ max(1, Δ_max e^{F_max}), not the printed Δ_max e^{F_max} max{1, 1/(1-e^{-Δ_min e^{F_min}})}. The printed expression can be unbounded when Δ_min e^{F_min} is small and does not describe the actual Lipschitz behavior of the loss. This error propagates into the Rademacher bound and the explicit constants in Theorem 8.3.
  4. [§5.1, Assumption 5.1 and §10, Theorems 10.1-10.2] The ranking-risk bound does not follow from the score definition. The score is f=α+β bS-τ d_m^2, i.e. it uses squared distance, but the proof inserts linear distortion inequalities into the score gap: f(pos)-f(neg) ≥ τ_r(α_m(θ+γ)-β_m(θ-γ)). For an embedding with distance distortion bounds α_m d_G ≤ d_m ≤ β_m d_G, the squared-distance gap would be (α_m(θ+γ))^2 - (β_m(θ-γ))^2, not the linear expression. Moreover, a lower bound on the score gap does not by itself yield the claimed bound on misranking probability. The theorem is unproved as stated.
  5. [§4, Proposition 4.1] All consistency and asymptotic-normality results depend on strict stationarity and β-mixing with Σ β(k)^{1/3}<∞. No argument or evidence is provided that temporal KG event streams satisfy this condition; Section 13 explicitly lists non-mixing and non-stationarity as failure modes but gives only diagnostic heuristics, not testable sufficient conditions. The claim 'consistency under temporal dependence' is therefore restricted to an unverified class of processes. This is a substantial limitation that should be stated prominently even if the other issues were fixed.
  6. [§4, Proposition 4.1] The claim that the MaxEnt solution is the minimum-norm parameter vector satisfying the moment constraints is not correct. The dual of the MaxEnt problem with constraints E_p[ϕ]=c is minimization of the log-partition A(w)-w·c over w, not minimization of ∥w∥ on the set {w: E_{p_w}[ϕ]=c}. The subsequent 'tightest norm-based bound' on Rademacher complexity is therefore not established. This is a secondary issue but affects the paper's advertised optimal-complexity control.
minor comments (6)
  1. [Appendix C] The notation table defines λ∈[0,1] as a 'mixing parameter', while Sections 3 and 6 use λ as the temperature in softmax/soft-min updates. This ambiguity should be fixed.
  2. [Algorithm 1] Step 3 updates embeddings by gradient descent on L_cll, but Step 2 solves a MaxEnt problem for fixed w; the relationship between the MaxEnt multipliers and the L_cll gradients is not specified. The algorithm would benefit from an explicit single objective or a proof that each step decreases it.
  3. [Theorem 3.4] The rate-distortion lower bound R(D) ≥ max{H_graph - log(1+D), H_temporal - log(1+D)} is asserted with only 'apply rate-distortion theory'. The proof is too sketchy to verify, and the max form is not derived from any stated source or lemma.
  4. [Theorem 3.2] The proof sketch asserts that J satisfies the Kurdyka-Łojasiewicz property, but J involves the cloglog loss, hyperbolic distances, and softmax terms; no argument is given that J is real analytic or semialgebraic on the relevant domain.
  5. [Example 9.2] The statement that Euclidean embedding of 2^k leaves requires distortion Ω(2^{k/d}) is not the Bourgain bound cited, and the subsequent substitution k=Ω(d log n) leading to Ω(n) is unclear. This example should be made rigorous or labeled heuristic.
  6. [Definition 5.3] The continuous-time intensity is written λ_{h,r,t}(s)=exp(f_s(h,r,t)), but f_s is not formally defined for continuous times; only the discrete score f_u is defined. Please clarify the extension.

Circularity Check

2 steps flagged · score 6.0 of 10

Canonical score is the exponent of the MaxEnt solution, i.e. the dual of the chosen moment constraints; Remark 4.1's denial of circularity is unsupported.

  1. self definitional [Definition 4.1 / Theorem 4.2 / Corollary 4.1]
    "The constraint set is: F_{h,r,u} := { p ∈ Δ_K : Σ_i p_i D_i = c^{(D)}_{r,u}, Σ_i p_i bS_i = c^{(S)}_{r,u} } ... has a unique solution of the form p*_i = exp{α* + β* bS_i − τ* d²_i}/Σ_j exp{α* + β* bS_j − τ* d²_j} ... The MaxEnt solution uniquely induces the canonical score function: f_u(h,r,t)=α_{r,u}+β_r bS_u(h,r,t)−τ_r d²(h,r,t;u)"

    The 'canonical score' f in (15) is exactly the exponent of the MaxEnt density in (10) up to the normalizer. That density is the generic exponential-family form forced by the chosen moment constraints (7)-(8), whose sufficient statistics are (1, bS, d²). Thus the result is the dual of the input constraints, not an independent first-principles prediction. Theorem 4.2 also fixes one head h and produces per-head multipliers; Corollary 4.1 silently replaces them with head-independent β_r, τ_r, an extra assumption not proved by the theorem. The advertised unique canonical score is therefore the chosen sufficient statistics renamed as an output.

  2. other [Remark 4.1 / Algorithm 1 Steps 2 and 5]
    "Circular dependencies are broken through the decoupled optimization scheme (Algorithm 1), while regularized feasibility conditions handle rank deficiency automatically. The framework is now mathematically complete without circular reasoning."

    Algorithm 1 does not supply the promised resolution: Step 2 solves the MaxEnt problem only for fixed mixture weights w, while Step 5 updates w from distortion energies that depend on the same embeddings and on the composite energy D used in the MaxEnt constraints. No fixed-point theorem in the paper shows that the converged parameters satisfy the original moment constraints (7)-(8); Theorem 3.2 only gives stationarity of the surrogate J. The paper's assertion that circular dependencies are 'broken' is thus unsupported, and the derivation of D, the score, and the mixture weights remains mutually dependent.

full rationale

The cloglog characterization (Theorems 5.1–5.2) and the generalization bounds are self-contained mathematical arguments and do not reduce to their inputs. The geometric necessity claims rely on external Bourgain/Gromov results, not on self-citation. However, the paper's central advertised result—the unique canonical score function—is the log of the MaxEnt distribution whose sufficient statistics are exactly the features (1, bS, d²) selected in the moment constraints. The Lagrange multipliers are fitted to the empirical moments, so presenting them as a unique prediction is a fitted-input/self-definitional move. The paper's own Remark 4.1 asserts without proof that the decoupled Algorithm 1 breaks circularity, but Algorithm 1 alternates between the MaxEnt problem and the mixture-weight update without a fixed-point theorem tying the converged solution to the moment constraints. The head-independence of β_r, τ_r in Corollary 4.1 is not established by the per-head Theorem 4.2; this unsupported inference strengthens rather than removes the circularity concern. No self-citation chain is present, so the score is 6 rather than higher.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

The framework depends on many fitted quantities (multipliers, embeddings, transports, mixture weights, hand-set regularizers) and on strong statistical assumptions (full rank, stationarity, beta-mixing, compactness). The paper provides no independent empirical or formal handle that would test these assumptions, so the central claims are not validated outside the paper's own construction.

free parameters (6)
  • MaxEnt multipliers alpha_{r,u}, beta_r, tau_r = optimized via alternating MaxEnt and gradient steps
    Determined from empirical moments and the cloglog likelihood, not from external benchmarks or constraints.
  • mixture weights w_{r,m} = softmax(-E_{r,m}/lambda)
    Fitted to distortion energies computed from the trained embeddings; the paper provides no independent source for these weights.
  • entity embeddings x_i^{(m)}(u) = gradient descent on cloglog loss
    All embedding coordinates are trainable parameters; the central score depends on them.
  • relation transports phi_r^{(m)} = gradient descent
    Translation/rotation parameters fitted per relation and per metric.
  • regularizer strengths and temperature schedule (lambda_gate, lambda_rad, lambda_0, T) = hand-set in Algorithm 3
    Chosen by hand with no theoretical selection rule; affects the mixture convergence and optimization.
  • embedding dimensions d_E, d_H, d_S = not specified
    Free architectural choices; the theoretical bounds depend on them through R_E, R_H, delta_S.
assumptions (7)
  • domain assumption The observed temporal graph process is strictly stationary and beta-mixing with sum beta(k)^{1/3} finite
    Assumption 5.1; used in Theorem 8.3 for effective sample size and in Theorems 10.1-10.2 for consistency and asymptotic normality.
  • ad hoc to paper The feature matrix [1, bS_i, d_i^2] has full rank 3 after possible adaptive augmentation
    Assumption 4.1; needed for unique Lagrange multipliers in Theorem 4.2, with no proof that sparse TKG candidate sets satisfy it.
  • ad hoc to paper The parameter space is compact by construction or through coercive regularization
    Assumption 3.1; underpins existence, convergence, and generalization arguments, but is not shown to hold during optimization.
  • domain assumption Independent increments and Poisson or Bernoulli coarsening for temporal events
    Definition 5.3 and Theorem 5.2 condition 1; this is essentially assuming a Poisson process, which yields the cloglog link.
  • standard math MaxEnt moment constraints are empirical moments computed from observed data and energies D
    Definition 4.1 and Theorem 4.1; standard exponential-family sufficient statistics reasoning.
  • standard math Canonical geodesic distances in Euclidean, hyperbolic, and spherical spaces
    Definition 3.1; standard geometry used without proof.
  • domain assumption Bourgain's Euclidean lower bound and Gromov's tree embeddings transfer to temporal knowledge graph scoring
    Sections 11 and Example 9.2 invoke these theorems for tree metrics and then apply them to TKG scoring, but no formal reduction is supplied.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MAGNET-KG: Maximum-Entropy Geometric Networks for Temporal Knowledge Graphs: Theoretical Foundations and Mathematical Framework." pith.science (2026). https://pith.science/paper/5PV26T43

@misc{pith2026250910587,
  author       = {Pith},
  title        = {Pith review of: MAGNET-KG: Maximum-Entropy Geometric Networks for Temporal Knowledge Graphs: Theoretical Foundations and Mathematical Framework},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5PV26T43}},
  note         = {Machine review of arXiv:2509.10587}
}
read the original abstract

We present a unified theoretical framework for temporal knowledge graphs grounded in maximum-entropy principles, differential geometry, and information theory. We prove a unique characterization of scoring functions via the maximum-entropy principle and establish necessity theorems for specific geometric choices. We further provide rigorous derivations of generalization bounds with explicit constants and outline conditions under which consistency guarantees hold under temporal dependence. The framework establishes principled foundations for temporal knowledge graph modeling with formal connections to differential geometric methods.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 2 linked inside Pith

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should not add it explicitly Type <Return> for now, but then later remove the command n...

  3. [3]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@first@sw \@firstoftwo \@ifundefined NAT@b*@#2 \@firstoftwo @num @NAT@ctr \@secondoft...

  4. [4]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibsetup #1 @NAT@ctr @ @openbib .11em \@plus.33em \@minus.07em 4000 4000 `\.\@m @bibit...

  5. [5]

    Lectures on functional equations and their applications, volume 19

    J \'a nos Acz \'e l. Lectures on functional equations and their applications, volume 19. Academic Press, 1966

    1966

  6. [6]

    Neural network learning: Theoretical foundations

    Martin Anthony and Peter L Bartlett. Neural network learning: Theoretical foundations. cambridge university press, 1999

    1999

  7. [7]

    Variational convergence for functions and operators

    Hedy Attouch. Variational convergence for functions and operators. Pitman Advanced Publishing Program, 1984

    1984

  8. [8]

    Information and exponential families: in statistical theory

    Ole Barndorff-Nielsen. Information and exponential families: in statistical theory. John Wiley & Sons, 1978

    1978

Show all 44 references
  1. [9]

    Rademacher and G aussian complexities: Risk bounds and structural results

    Peter L Bartlett and Shahar Mendelson. Rademacher and G aussian complexities: Risk bounds and structural results. Journal of Machine Learning Research, 3 0 (Nov): 0 463--482, 2002

    2002

  2. [10]

    Rate distortion theory: A mathematical basis for data compression

    Toby Berger. Rate distortion theory: A mathematical basis for data compression. 1971

    1971

  3. [11]

    Concentration inequalities: A nonasymptotic theory of independence

    St \'e phane Boucheron, G \'a bor Lugosi, and Pascal Massart. Concentration inequalities: A nonasymptotic theory of independence. Oxford university press, 2013

    2013

  4. [12]

    On L ipschitz embedding of finite metric spaces in H ilbert space

    Jean Bourgain. On L ipschitz embedding of finite metric spaces in H ilbert space. Israel Journal of Mathematics, 52 0 (1-2): 0 46--52, 1985

    1985

  5. [13]

    Convex optimization

    Stephen Boyd and Lieven Vandenberghe. Convex optimization. Cambridge university press, 2004

    2004

  6. [14]

    Introduction to strong mixing conditions, volume 1

    Richard C Bradley. Introduction to strong mixing conditions, volume 1. Kendrick Press, 2007

    2007

  7. [15]

    Hyperbolic graph convolutional neural networks

    Ines Chami, Rex Ying, Christopher R \'e , and Jure Leskovec. Hyperbolic graph convolutional neural networks. Advances in neural information processing systems, 32, 2019

    2019

  8. [16]

    Refh: A reference-based framework for hyperbolic knowledge graph embeddings

    Ines Chami, Adva Gu, Lior Wolf, and Christopher R \'e . Refh: A reference-based framework for hyperbolic knowledge graph embeddings. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp.\ 4302--4313, 2020

    2020

  9. [17]

    Elements of information theory

    Thomas M Cover and Joy A Thomas. Elements of information theory. John Wiley & Sons, 2012

    2012

  10. [18]

    Analysis of survival data, volume 21

    David R Cox and David Oakes. Analysis of survival data, volume 21. CRC Press, 1984

    1984

  11. [19]

    An introduction to the theory of point processes: volume I : elementary theory and methods

    Daryl J Daley and David Vere-Jones. An introduction to the theory of point processes: volume I : elementary theory and methods . Springer Science & Business Media, 2003

    2003

  12. [20]

    Differential geometry of curves and surfaces, volume 2

    Manfredo Perdigao Do Carmo. Differential geometry of curves and surfaces, volume 2. Prentice-hall Englewood Cliffs, 1992

    1992

  13. [21]

    Hyperbolic groups

    Mikhael Gromov. Hyperbolic groups. Essays in group theory, 8: 0 75--263, 1987

    1987

  14. [22]

    The minimum description length principle

    Peter D Gr \"u nwald. The minimum description length principle. MIT press, 2007

    2007

  15. [23]

    Dyre: A dynamic reasoning network for temporal knowledge graph completion

    Jia Han, Deva Ramanan, and Tian-tian Ma. Dyre: A dynamic reasoning network for temporal knowledge graph completion. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: Findings, pp.\ 2509--2519, 2020

    2020

  16. [24]

    Chronor: A time-aware embedding model for temporal knowledge graph completion

    Jia Han, Tian-tian Ma, and Deva Ramanan. Chronor: A time-aware embedding model for temporal knowledge graph completion. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pp.\ 8779--8787, 2021 a

    2021

  17. [25]

    Roth: Relation-wise rotation for hyperbolic knowledge graph embedding

    Xiao Han, Zhi-Chao Liu, Maosong Sun, Yang Liu, and Huan-Bo Luan. Roth: Relation-wise rotation for hyperbolic knowledge graph embedding. In Proceedings of the Web Conference 2021, pp.\ 1135--1146, 2021 b

    2021

  18. [26]

    Spectra of some self-exciting and mutually exciting point processes

    Alan G Hawkes. Spectra of some self-exciting and mutually exciting point processes. Biometrika, 58 0 (1): 0 83--90, 1971

    1971

  19. [27]

    Probability theory: the logic of science

    Edwin T Jaynes. Probability theory: the logic of science. Cambridge university press, 2003

    2003

  20. [28]

    Recurrent event network for reasoning over temporal knowledge graphs

    Woojeong Jin, Meng Qu, Xisen Jin, and Jiliang Tang. Recurrent event network for reasoning over temporal knowledge graphs. arXiv preprint arXiv:2004.05537, 2020

    2004 arXiv

  21. [29]

    Tensor-based temporal knowledge graph completion

    Timoth \'e e Lacroix, Guillaume Obozinski, and Nicolas Usunier. Tensor-based temporal knowledge graph completion. arXiv preprint arXiv:2004.09142, 2020

    2004 arXiv

  22. [30]

    Introduction to R iemannian manifolds , volume 176

    John M Lee. Introduction to R iemannian manifolds , volume 176. Springer, 2018

    2018

  23. [31]

    xerte: a cross-modal entity and relation type enhanced model for temporal knowledge graph completion

    Yucheng Li, Tao Ge, Dan Zhang, Siyuan Wang, and Baobao Chang. xerte: a cross-modal entity and relation type enhanced model for temporal knowledge graph completion. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp.\ 8861--8871, 2021

    2021

  24. [32]

    On the method of bounded differences

    Colin McDiarmid. On the method of bounded differences. Surveys in combinatorics, 141 0 (1): 0 148--188, 1989

    1989

  25. [33]

    Foundations of machine learning

    Mehryar Mohri, Afshin Rostamizadeh, and Ameet Talwalkar. Foundations of machine learning. MIT press, 2018

    2018

  26. [34]

    Poincar \'e embeddings for learning hierarchical representations

    Maximillian Nickel and Douwe Kiela. Poincar \'e embeddings for learning hierarchical representations. In Advances in neural information processing systems, pp.\ 6338--6347, 2017

    2017

  27. [35]

    A generalization of the probit and logit methods for dose response curves

    Ross L Prentice. A generalization of the probit and logit methods for dose response curves. Biometrics, pp.\ 761--768, 1976

    1976

  28. [36]

    Convex analysis

    R Tyrrell Rockafellar. Convex analysis. Princeton university press, 1970

    1970

  29. [37]

    Variational analysis, volume 317

    R Tyrrell Rockafellar and Roger J-B Wets. Variational analysis, volume 317. Springer Science & Business Media, 2009

    2009

  30. [38]

    On the density of families of sets

    Norbert Sauer. On the density of families of sets. Journal of Combinatorial Theory, Series A, 13 0 (1): 0 145--147, 1972

    1972

  31. [39]

    A mathematical theory of communication

    Claude E Shannon. A mathematical theory of communication. The Bell system technical journal, 27 0 (3): 0 379--423, 1948

    1948

  32. [40]

    Rotate: Knowledge graph embedding by relational rotation in complex space

    Zhiqing Sun, Zhi-Hong Deng, Jian-Yun Nie, and Jian Tang. Rotate: Knowledge graph embedding by relational rotation in complex space. In International Conference on Learning Representations, 2019

    2019

  33. [41]

    Asymptotic statistics, volume 3

    Aad W Van der Vaart. Asymptotic statistics, volume 3. Cambridge university press, 2000

    2000

  34. [42]

    Statistical learning theory, volume 1

    Vladimir Vapnik. Statistical learning theory, volume 1. Wiley New York, 1998

    1998

  35. [43]

    Maximum likelihood estimation of misspecified models

    Halbert White. Maximum likelihood estimation of misspecified models. Econometrica: Journal of the econometric society, pp.\ 1--25, 1982

    1982

  36. [44]

    Rates of convergence for empirical processes of stationary mixing sequences

    Bin Yu. Rates of convergence for empirical processes of stationary mixing sequences. The Annals of Probability, pp.\ 94--116, 1994

    1994

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.