Pith. sign in

REVIEW 4 major objections 6 minor 37 references

Mirror Descent Using the Tempesta Generalized Multi-parametric Logarithms

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A multiparametric deformed logarithm generates a tunable family of mirror descent updates that specializes to previously studied deformed-logarithm updates.

desk verdict A genuine but unvalidated extension of mirror descent to Tempesta logarithms; the stress-test's normalization objection is wrong, yet the paper's real gaps are missing proofs, missing error control, and zero experiments. read the letter →

arxiv 2506.13984 v1 pith:BIPAZBEK submitted 2025-06-08 stat.ML cs.AIcs.LG

Andrzej Cichocki This is my paper
classification stat.MLcs.AIcs.LG MSC 90C2568W2794A17
keywords mirrordescentTempestalogarithmdeformedlogarithmsBregmandivergenceexponentiatedgradientgeneralizedentropymirrorlesshyperparameterlearning
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 proposes a family of mirror descent algorithms whose link function is the Tempesta multiparametric logarithm $\log_{\phi,\alpha,\sigma}^{\mathrm{Te}}(x)$, a deformed logarithm controlled by a generating function $\phi$ and two scalar parameters $\alpha,\sigma$. The intended payoff is a theoretically infinite class of exponentiated-gradient-style updates that interpolates among and generalizes the Tsallis, Kaniadakis, Euler, and HTG logarithms already used in mirror descent. Because the shape of the logarithm can be tuned by hyperparameters, the paper argues that the updates can adapt to the geometry or distribution of the training data, and that the hyperparameters themselves can be learned. The two concrete deliverables are a generalized multiplicative mirror descent update and an additive mirror-less update, both followed by unit-simplex normalization.

What carries the argument

The load-bearing object is the Tempesta functional multiparametric logarithm $\log_{\phi,\alpha,\sigma}^{\mathrm{Te}}(x)$, a deformed logarithm built from a generating function $\phi$ and parameters $\alpha,\sigma$, which reduces to $\ln(x)$ when $\sigma=0$ and to the Tsallis $q$-logarithm when $\sigma=q-1$. It plays the role of the mirror link $f=\nabla F$; the paper uses the Lagrange–Bürmann inversion theorem to approximate its inverse, the generalized exponential $\exp_{\phi,\alpha,\sigma}^{\mathrm{Te}}$, and composes the two through the generalized multiplication $x\otimes_G \exp_T^{\mathrm{Te}}(y)=\exp_T^{\mathrm{Te}}(\log_T^{\mathrm{Te}}(x)+y)$. The Bregman divergence from Eq. (2) and the normalized loss $\widehat{L}(w)=L(w/\|w\|_1)$ carry the optimization argument.

What would settle it

Scan the admissible parameter region of Eq. (6), for instance $\alpha\in[-2,2]$, $\sigma\in[-2,2]$, and $\phi(x)=x^r$ on a grid of $r$, and numerically evaluate $\frac{d}{dx}\log_T(x)$ at $x=1$ together with the sign of $\frac{d^2}{dx^2}\log_T(x)$ over $x>0$. A single parameter triple that satisfies Eq. (6) but violates the unit derivative, or that shows a positive second derivative anywhere, refutes the claim that the Tempesta logarithm is a valid mirror link.

Watch

Extended reading notes

Core claim

The central claim is that replacing the standard logarithm in mirror descent with the Tempesta functional multiparametric logarithm gives valid mirror descent updates. On the paper's own terms, this means treating $\log_{\phi,\alpha,\sigma}^{\mathrm{Te}}$ as the link function $f=\nabla F$, with $F(w)=\int \log_T(w)\,dw$ strictly convex, and using the inverse generalized exponential in the implicit prox update. The result is $w_{t+1}=\exp_{\phi,\alpha,\sigma}^{\mathrm{Te}}(\log_{\phi,\alpha,\sigma}^{\mathrm{Te}}(w_t)-\eta_t\nabla \widehat{L}(w_t))$, followed by $\ell^1$-normalization (Eq. 48), plus a mirror-less additive variant (Eq. 51). Since the exact inverse of the Tempesta logarithm is generally not available in closed form, the paper approximates the inverse exponential by a Lagrange–Bürmann power series around $1$. The paper further shows that special parameter choices reproduce the Tsallis, Kaniadakis, Euler, and HTG logarithms and their exponentials, so the proposed updates form one parametrized family that contains several previously studied algorithms.

Load-bearing premise

The load-bearing premise is that the Tempesta logarithm is a valid mirror link for every allowed parameter choice: strictly increasing, concave, vanishing at $x=1$ with unit derivative, so that $F(w)=\int \log_T(w)\,dw$ is strictly convex and the Bregman divergence and inverse-exponential updates are well defined.

Editorial extensions

If this is right

  • For each admissible choice of $(\phi,\alpha,\sigma)$, Eqs. (48) and (51) define concrete algorithms on the unit simplex, so the paper yields a broad family of implementable updates rather than a single method.
  • Because Tsallis, Kaniadakis, Euler, and HTG updates are special cases, results and intuition for those algorithms transfer into the new parametrized family.
  • Tuning or learning the hyperparameters changes the Bregman geometry, so the method can in principle match the geometry of the data distribution, which the paper argues can improve convergence.
  • The mirror-less additive update (51) provides a primal-only variant that avoids explicit inversion of the deformed logarithm, at the cost of a gradient scaling by the inverse diagonal of the link function's Jacobian.
  • The family offers flexible regularization for sparse-data problems, with potential applications the paper lists in deep learning, classification, clustering, prediction, and online portfolio selection.

Reading between the lines

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

  • Beyond the paper, a natural next step is an empirical study that learns $(\alpha,\sigma,\phi)$ on online-learning or portfolio-selection benchmarks and compares the learned geometry against fixed Tsallis, Kaniadakis, and Euler updates.
  • If the derivative of the Tempesta logarithm at $x=1$ is not actually $1$ for generic admissible parameters, the exponential-log composition in Eq. (48) would not reduce to gradient descent at small step size; reparameterizing the family to enforce this normalization would preserve the mirror-descent interpretation.
  • The mirror-less update (51) can be read as a diagonal natural-gradient step under the Bregman metric; the paper notes this connection, and proving convergence rate bounds for the full parametrized family is an extension beyond the paper's scope.
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.

Referee Report

4 major / 6 minor

Summary. The paper proposes using Tempesta's multi-parametric generalized logarithm log_T as the link function in mirror descent (MD) and mirror-less mirror descent (MMD). The central formulas are Eq. (48), the multiplicative update w_{t+1} = exp_T(log_T(w_t) - η_t ∇\hat{L}(w_t)) followed by unit-simplex normalization, and Eq. (51), the additive MMD update. The G-multiplication in Eq. (49) is defined so that the multiplicative form holds by construction. The paper surveys special cases (Tsallis, Kaniadakis, Euler, HTG), derives an approximate inverse exponential via Lagrange inversion in Eq. (40), and claims that the hyperparameters can be learned to adapt to data. A note on the appended stress-test: the objection that Eq. (5) fails the normalization f'(1)=1 does not land, because direct differentiation of Eq. (5) gives f'(1)=1; the substantive problems are elsewhere, notably the expansion coefficients in Eq. (8), the unproved domain conditions, and the absence of any convergence analysis or experiments.

Significance. If fully worked out, the construction would give a parametric family of exponentiated-gradient-style algorithms that contains several known deformed-logarithm algorithms as special cases, and the connection to trace-form entropies is interesting. The paper is explicit about the special cases and the standard Lagrange-inversion route to the inverse link. However, the central algorithmic contribution is essentially a substitution of a parametric link function into the textbook MD update: Eq. (48) is just Eq. (3) with f = log_T, and Eq. (49) defines the G-product so that the identity holds. No convergence theorem, no numerical experiments, and no hyperparameter-learning procedure are provided, and the only non-definitional mathematical step, the expansion in Eq. (8), appears to be inconsistent with Eq. (5). Thus the paper is currently a proposal with a substantial correctness gap and no validation.

major comments (4)
  1. [Section II, Eqs. (5), (8)-(9); Section III, Eq. (40)] The Taylor expansion of log_T in powers of ln x does not match Eq. (5). For the affine generating function φ(x)=a x + c, which gives the Tsallis q-logarithm, Eq. (5) reduces to log_T(x) = (1 - x^{-σ})/σ. With t = ln x this has expansion t - (σ/2)t^2 + (σ^2/6)t^3 + ⋯, so in the notation of Eq. (9) we must have a1 = -σ and a2 = σ^2. Substituting φ''=φ'''=0 into Eq. (8) instead gives a1 = -1, independent of σ. More generally, differentiating Eq. (5) yields a1 = -σ + σ α^2 φ''(α)/(1 - αφ'(α)), which is not the expression implied by Eq. (8). Since the inverse series in Eq. (40) and hence the approximate exponential used in update (48) are built from these coefficients, this is a load-bearing error that must be corrected and the derivation supplied.
  2. [Section II, Eq. (6) and the property list after Eq. (9)] The manuscript asserts that condition (6) and an unspecified 'suitable chosen range' of α, σ, φ guarantee monotonicity, concavity, log_T(1)=0, log_T'(1)=1, and strict convexity of F in Eq. (46). These properties are load-bearing because the Bregman divergence (2) and the invertibility used in Eq. (48) require log_T to be a strictly increasing bijection from R_+ to its range and F to be strictly convex. The paper gives no proof and no explicit admissible parameter ranges, and it does not state precisely which theorem of Refs. [31,32] establishes these facts. The discussion should either prove the needed properties or state the exact theorem and parameter restrictions. Relatedly, the statement in Section I that the link function must be strictly concave is not the correct general condition for a Bregman divergence: strict convexity of F requires f to be strictly increasing, not concave, and the paper should clarify this.
  3. [Sections IV and V] No convergence analysis, no numerical experiments, and no hyperparameter-learning procedure are presented. The abstract and Section V claim that hyperparameters can be learned to adapt to the geometry of the data and that the algorithms can achieve 'much faster convergence rates', but the body of the paper only writes down updates (48) and (51). For an algorithms paper, the central practical claims are unsupported by either theory or evidence. A convergence theorem under explicit conditions on the link function, or a set of experiments illustrating the behavior of the proposed family, is needed before the contribution can be evaluated.
  4. [Section IV, Eqs. (48)-(49)] The 'generalized multiplicative update' in Eq. (48) is a restatement of the standard mirror descent update (3) once the link function is set to log_T. Equation (49) defines the G-multiplication so that the equality in (48) holds by construction, rather than deriving a new algorithmic rule. The paper should clarify what genuinely new algorithm is being proposed beyond plugging the Tempesta logarithm into the standard MD template, and it should distinguish this definitional aspect from the separate question of whether the resulting mirror map has useful properties.
minor comments (6)
  1. [Eq. (1)] The constraint is malformed: '||w||_1 = Σ_i, w_i = 1' should be written as a single condition such as ||w||_1 = Σ_i w_i = 1.
  2. [Eq. (5)] The denominator '1−αϕ′(α' has an unclosed parenthesis and the multi-line display is difficult to parse; the formula should be typeset unambiguously.
  3. [Eq. (40)] The displayed expansion contains typographical inconsistencies, including '1−a−b' in the second line and an unmatched parenthesis near 'a_2)'; the inverse series should be re-derived and typeset carefully.
  4. [Section I and throughout] The paper alternates between 'MDD' and 'MD' without defining the former acronym; please standardize the terminology.
  5. [References] Reference [10] is missing complete publication details, and reference [35] misspells 'Tsallis' as 'Tasllis'; please correct these.
  6. [Section V] The word 'predication' should be 'prediction' in the last paragraph.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the MD updates are direct substitutions into the standard mirror-descent template, and the self-citations are not load-bearing.

full rationale

The paper's central update, Eq. (48), is obtained by substituting the Tempesta logarithm into the general mirror-descent formula (3); the second equality in (48) merely re-expresses this substitution using the G-multiplication defined in Eq. (49). That definitional identity is not a fitted prediction or an input-output circularity; it is notation introduced at the point of use. The inverse series in Eq. (40) is derived from the stated power series (8)-(9) by the Lagrange-Bürmann inversion theorem, an external mathematical result, so no fitted parameter is renamed as a prediction. The hyperparameter-learning claim in the abstract and conclusions is aspirational rather than a fitted-then-predicted experiment. The self-citations [10]-[12] are used for contextual prior work and for the normalized-loss implementation detail (50), but the central derivation does not depend on an unverified claim from those references. The main weakness is a correctness risk, not circularity: Section II asserts, without proof, that the Tempesta logarithm satisfies f(1)=0, f'(1)=1, strict concavity, and strict convexity of F in Eq. (46), and Eq. (5) may not satisfy f'(1)=1 for generic parameters. This is an unsupported or internally inconsistent premise, and it would undermine validity of the Bregman divergence and the update, but it is not a self-referential reduction. Consequently, no circular step meets the evidentiary bar; the paper is best characterized as an application of an externally defined function family to a standard algorithm template, with the relevant mathematical caveats flagged as correctness issues rather than circularity.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The paper leans on Tempesta's existence theorem for the logarithm family, on Lagrange inversion, and on unproven regularity properties of the log it introduces. It introduces no new physical entities; the G-multiplication operator in Eq. (49) is a notational definition rather than an invented entity. The main free parameters are the log's hyperparameters, which are not fitted in the paper.

free parameters (1)
  • Tempesta hyperparameters alpha, sigma, and generating function phi = None provided; the paper says they can be learned but gives no fitting procedure
    The proposed MD updates depend on these free parameters. The paper claims they can be tuned to data, but no values or learning rule are supplied, so they remain free parameters of the method.
assumptions (4)
  • standard math Lagrange inversion theorem is valid and the truncated power series gives a usable inverse exponential.
    Used in Section III, Eq. (40), to estimate the inverse of the Tempesta logarithm. The theorem is standard, but the paper does not bound the truncation error.
  • domain assumption F(w) = integral of log_T(w) dw is strictly convex and the induced Bregman divergence is valid for the chosen parameter ranges.
    Required for the MD and mirror-less updates in Section IV. The paper asserts convexity in Eq. (46) without proof, and the associated normalization claim is questionable.
  • ad hoc to paper Condition (6) on phi and the hyperparameters ensures strict concavity and monotonicity of the Tempesta logarithm.
    This condition is stated in Section II without derivation. It is needed for the log to behave as a valid link function, but no proof is given that the condition holds for any non-trivial parameter choice.
  • domain assumption Standard mirror-descent convergence guarantees transfer to the new multi-parametric link function.
    The conclusions claim faster convergence and better flexibility, but the paper provides no convergence analysis, implicitly relying on known MD theory without verifying its assumptions for this link.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mirror Descent Using the Tempesta Generalized Multi-parametric Logarithms." pith.science (2026). https://pith.science/paper/BIPAZBEK

@misc{pith2026250613984,
  author       = {Pith},
  title        = {Pith review of: Mirror Descent Using the Tempesta Generalized Multi-parametric Logarithms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BIPAZBEK}},
  note         = {Machine review of arXiv:2506.13984}
}
read the original abstract

In this paper, we develop a wide class Mirror Descent (MD) algorithms, which play a key role in machine learning. For this purpose we formulated the constrained optimization problem, in which we exploits the Bregman divergence with the Tempesta multi-parametric deformation logarithm as a link function. This link function called also mirror function defines the mapping between the primal and dual spaces and is associated with a very-wide (in fact, theoretically infinite) class of generalized trace-form entropies. In order to derive novel MD updates, we estimate generalized exponential function, which closely approximates the inverse of the multi-parametric Tempesta generalized logarithm. The shape and properties of the Tempesta logarithm and its inverse-deformed exponential functions can be tuned by several hyperparameters. By learning these hyperparameters, we can adapt to distribution or geometry of training data, and we can adjust them to achieve desired properties of MD algorithms. The concept of applying multi-parametric logarithms allow us to generate a new wide and flexible family of MD and mirror-less MD updates.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 34 canonical work pages

  1. [1]

    Amari, S. (2016). Information Geometry and its Application.Springer, Berlin

    work page 2016

  2. [2]

    Amari, S. (2009). Alpha-divergence is unique, belonging to both f-divergence and Bregman divergence classes.IEEE Transactions on Informations Theory,55, 4925–4931

    work page 2009

  3. [3]

    Amid, E., and Warmuth, M. K. (2020). Reparameterizing mirror descent as gradient descent. InProceedings of the 34th International Conference on Neural Information Processing Systems (NIPS’20), Curran Associates Inc., Red Hook, NY , USA, Article 706, 8430-8439

    work page 2020

  4. [4]

    and Warmuth, M

    Amid, E. and Warmuth, M. K. (2020). Winnowing with Gradient Descent. InProceedings of the 33rd International Conference on Algorithmic Learning Theory, PMLR 125:163-182

    work page 2020

  5. [5]

    and Thakurta, A

    Amid, E., Ganesh, A., Mathews, R., Ramaswamy, S., Song, S., Steinke, T., Suriyakumar, V .M., Thakkar, O. and Thakurta, A. (2022). Public data-assisted mirror descent for private model training. InInternational Conference on Machine Learning, (pp. 517-535). PMLR

    work page 2022

  6. [6]

    and Hassibi, B

    Azizan, N., Lale, S. and Hassibi, B. (2021). Stochastic mirror descent on overparameterized nonlinear models.IEEE Transactions on Neural Networks and Learning Systems, 33(12), pp.7717-7727

    work page 2021

  7. [7]

    and Teboulle, M

    Beck, A. and Teboulle, M. (2003). Mirror descent and nonlinear projected subgradient methods for convex optimization.Operations Research Letters, 31(3), pp.167-175

    work page 2003

  8. [8]

    Borges and I

    E.P. Borges and I. Roditi. (1998). A family of nonextensive entropies.Physics Letters A, 246(5):399–402

    work page 1998

Show all 37 references
  1. [9]

    Bregman, L. (1967). The relaxation method of finding a common point of convex sets and its application to the solution of problems in convex programming.Comp. Math. Phys., USSR, (7), pp. 200–217

    1967

  2. [10]

    Cichocki, A., Cruces, S., Sarmineto A., Tanaka T. (2024). Generalized Exponentiated Gradient Algorithms and Their Application to On-Line Portfolio Selection.IEEE Access, https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=10807168

    2024

  3. [11]

    Cichocki A. (2025). Generalized Exponentiated Gradient Algorithms Using the Euler Two-Parameter Logarithm.arXiv preprintarXiv:2502.17500

    2025 arXiv

  4. [12]

    Cichocki A., Tanaka T., Cruces S. (2025). Mirror Descent and Novel Exponentiated Gradient Algorithms Using Trace-Form Entropies and Deformed Logarithms.arXiv preprintarXiv:2503.08748

    2025

  5. [13]

    and Cruces, S

    Cichocki, A. and Cruces, S. and Amari, S. I. (2011). Generalized alpha-beta divergences and their application to robust nonnegative matrix factorization. Entropy, 13(1), pp. 134-170

    2011

  6. [14]

    and Singer, Y

    Ghai, U., Hazan, E. and Singer, Y . (2020). Exponentiated Gradient Meets Gradient Descent.In Proceedings of the 31st International Conference on Algorithmic Learning Theory, PMLR 117:386-407. https://doi.org/10.48550/arXiv.1902.01903

    2020 doi arXiv

  7. [15]

    and Srebro, N., (2021)

    Gunasekar, S., Woodworth, B. and Srebro, N., (2021). Mirrorless mirror descent: A natural derivation of mirror descent.International Conference on Artificial Intelligence and Statistics(pp. 2305-2313), PMLR

    2021

  8. [16]

    and Zhang, B.T

    Han, D.S., Kim, H., Lee, H., Ryu, J. and Zhang, B.T. (2022). Robust imitation via mirror descent inverse reinforcement learning.Advances in Neural Information Processing Systems, 35, pp.30031-30043

    2022

  9. [17]

    and Zhang, B.T., 2025)

    Han, D.S., Kim, J., Yoo, H.B. and Zhang, B.T., 2025). Variational Online Mirror Descent for Robust Learning in Schrödinger Bridge.arXiv preprint arXiv:2504.02618

    2025

  10. [18]

    and Gell-Mann, M., (2012)

    Hanel, R., Thurner, S. and Gell-Mann, M., (2012). Generalized entropies and logarithms and their duality relations.Proceedings of the National Academy of Sciences109(47), pp.19151-19154

    2012

  11. [19]

    and Scarfone, A.M., (2004)

    Kaniadakis, G., Lissia, M. and Scarfone, A.M., (2004). Deformed logarithms and entropies.Physica A: Statistical Mechanics and its Applications, 340(1-3), pp.41-49

    2004

  12. [20]

    Kaniadakis, M

    G. Kaniadakis, M. Lissia, and A.M. Scarfone. (2005). Two-parameter deformations of logarithm, exponential, and entropy: A consistent framework for generalized statistical mechanics.Physical Review E, 71(4):046128

    2005

  13. [21]

    Kaniadakis, G. (2009). Maximum entropy principle and power-law tailed distributions.Physical Journal B, (70), pp.3-13

    2009

  14. [22]

    and Wada, T., (2017)

    Kaniadakis, G., Scarfone, A.M., Sparavigna, A. and Wada, T., (2017). Composition law ofκ-entropy for statistically independent systems.Physical Review E, 95(5), p.052112

    2017

  15. [23]

    and Warmuth, M

    Kivinen, J. and Warmuth, M. K. (1997). Exponentiated Gradient versus Gradient Descent for Linear Predictors.Information and Computation, 132:1-63. http://dx.doi.org/10.1006/inco.1996.2612

    1997

  16. [24]

    Kivinen, J., Warmuth, M. K. (1995).Additive versus exponentiated gradient updates for linear prediction. In Proceedings of the Twenty-seventh Annual ACM Symposium on Theory of Computing (pp. 209-218). https://doi.org/10.1145/225058.225121 8

    1995

  17. [25]

    and Warmuth, M

    Majidi, N., Amid, E., Talebi, H. and Warmuth, M. K. (2021). Exponentiated Gradient Reweighting for Robust Training Under Label Noise and Beyond. ArXiv preprintarXiv:2104.01493

    2021 arXiv

  18. [26]

    D.P. Mittal. (1975). On some functional equations concerning entropy, directed divergence and inaccuracy.Metrika, 22(1):35–45

    1975

  19. [27]

    Nemirovsky and D

    A. Nemirovsky and D. Yudin. (1983). Problem Complexity and Method Efficiency in Optimization.John Wiley and Sons, https://doi.org/10.1137/1027074

    1983 doi

  20. [28]

    and Wada, T

    Scarfone, A.M., Suyari, H. and Wada, T. (2009). Gauss law of error revisited in the framework of Sharma-Taneja-Mittal information measure.Central European Journal of Physics, 7, pp.414-420

    2009

  21. [29]

    Shalev-Shwartz, S. (2011). Online learning and online convex optimization.Foundations and Trends in Machine Learning, 4(2):107-194

    2011

  22. [30]

    Sharma and I.J

    B.D. Sharma and I.J. Taneja. (1975). Entropy of type (α,β) and other generalized measures in information theory.Metrika, 22(1):205–215

    1975

  23. [31]

    Tempesta, P. (2015). A theorem on the existence of trace-form generalized entropies.Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 471(2183), p.20150165

    2015

  24. [32]

    Tempesta, P. (2016). Formal groups and Z-entropies.Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 472(2195), 20160143

    2016

  25. [33]

    I.J. Taneja. (2001). Generalized information measures and their applications. on-line book.URL www. mtm. ufsc. br/taneja/book/book. html

    2001

  26. [34]

    C. Tsallis. (1988). Possible generalization of Boltzmann-Gibbs statistics.Journal of statistical physics, 52(1):479–487

    1988

  27. [35]

    Tasllis, C. (1994). What are the numbers that experiments provide.Quimica Nova, 17,6, 468–471

    1994

  28. [36]

    Wada and A.M

    T. Wada and A.M. Scarfone. (2010). Finite difference and averaging operators in generalized entropies. InJournal of Physics: Conference Series, volume 201, page 012005. IOP Publishing

    2010

  29. [37]

    and Klabjan, D., (2024)

    Wang, S. and Klabjan, D., (2024). A Mirror Descent Perspective of Smoothed Sign Descent.arXiv preprintarXiv:2410.14158

    2024 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.