arxiv: 2605.08559 · v1 · submitted 2026-05-08 · 🧮 math.FA · cs.LG· cs.NA· cs.NE· math.NA· math.OC

Recognition: 1 theorem link

· Lean Theorem

Structure-Preserving Reconstruction of Convex Lipschitz Functionals on Hilbert Spaces from Finite Samples

Anastasis Kratsios

Pith reviewed 2026-05-12 01:07 UTC · model grok-4.3

classification 🧮 math.FA cs.LGcs.NAcs.NEmath.NAmath.OC

keywords convex functionalsLipschitz functionalsHilbert spacesReLU networksneural functionalsfunction reconstructionapproximation theorymachine learning

0 comments

The pith

Any convex L-Lipschitz functional on a compact convex set in a Hilbert space can be reconstructed to arbitrary uniform accuracy from finite point evaluations while exactly preserving convexity and the Lipschitz constant.

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

The paper demonstrates that convex functionals, which appear in risk measures, value functions, and machine learning losses, can be recovered from limited exact observations without losing their defining properties. Specifically, for any such functional on a compact convex domain in a Hilbert space, an explicit approximation formula exists that matches the original up to any small error, uses only finite linear measurements, and maintains exact convexity and Lipschitz behavior. This formula is realizable as a standard ReLU neural network. The work introduces convex neural functionals as an architecture class that enforces these properties for all parameters, offering a reliable way to learn structured functionals from data.

Core claim

For every compact convex set C in the Hilbert space H, every Lipschitz convex functional with constant L on C, and every positive epsilon, an explicit reconstruction exists that is convex, L-Lipschitz, and within epsilon of the functional on C. This reconstruction relies solely on finitely many linear measurements with vectors from a finite-dimensional subspace and can be implemented exactly by a ReLU multilayer perceptron. The paper defines convex neural functionals as a trainable class containing this reconstruction, ensuring convexity and Lipschitz continuity for every parameter choice.

What carries the argument

The structure-preserving reconstruction formula based on finite linear measurements in a finite-dimensional subspace, exactly implemented by a ReLU multilayer perceptron.

If this is right

Reconstructed functionals can be used directly in optimization problems where convexity ensures global optimality.
Risk measures and super-hedging prices can be approximated from data while retaining their mathematical properties.
Loss functions in machine learning can be learned with guaranteed convexity and Lipschitz regularity.
The convex neural functionals class provides a principled basis for training models without additional regularization for structure.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

This reconstruction technique might be adapted for cases with noisy data by incorporating robustness into the formula.
Connections to other approximation theories could lead to similar results for functionals with different regularity properties.
In computational practice, this could reduce the data requirements for learning convex models in high-dimensional spaces.

Load-bearing premise

The functional is available only through exact evaluations at a finite number of points on its compact convex domain.

What would settle it

Demonstrating a convex L-Lipschitz functional on a compact convex set for which there exists an epsilon such that no finite collection of its point values admits a convex L-Lipschitz approximant that is epsilon-close uniformly on the set.

Figures

Figures reproduced from arXiv: 2605.08559 by Anastasis Kratsios.

**Figure 1.** Figure 1: Max-pooling (cf. (2.2)) as upper envelope. Each partition group collects several PReLUactivated pre-pooling neurons hℓ,i and returns their pointwise maximum x pℓ`1q k “ maxiPrrkssΠpℓq hℓ,i. Since the pointwise maximum of convex functions is convex, every output neuron of the max-pooling layer is automatically convex. 3 Main Results We operate in the following setting. Setting 1. Fix L ą 0. Let H be a nonz… view at source ↗

**Figure 2.** Figure 2: Dual-ball discretization and reconstruction formula. a The η-net tpmu of V2 X BHp0, Lq discretizes the supremum in the dual form (5.9); each pm fixes the slope of one affine coordinate x p0q m “ xpm, xyH ` qm in Definition 2.2. b On the slice x “ pt, 0q, these affine pieces appear as faint coloured lines, while their pointwise maximum gives fε,d in (3.1). Convexity comes from the max-envelope structure, an… view at source ↗

**Figure 3.** Figure 3: CNF approximation of a randomly generated convex ReLU-MLP target in dimension din “ 1. The target network has two hidden layers of width 500 and 252,001 parameters, while the CNF has 81,402 trainable parameters, corresponding to a parameter ratio of 0.323. The CNF is trained on 1000 samples for 200 optimization iterations. Since all CNF iterates remain in the admissible architecture class, convexity is pre… view at source ↗

read the original abstract

Convex functionals are ubiquitous in applied analysis, appearing as value functions, risk measures, super-hedging prices, and loss functionals in machine learning. In many applications, however, the functional is only observed through finitely many exact pointwise evaluations. We ask whether a convex functional on a separable Hilbert space $H$ can be reconstructed, up to arbitrary uniform accuracy, by an explicit formula which preserves convexity and Lipschitz regularity and is finitely computable. We answer this affirmatively. For every compact convex $C\subseteq H$, every $L$-Lipschitz convex functional $\rho:C\to\mathbb{R}$, and every $\varepsilon>0$, we construct an explicit finite-sample reconstruction which is convex, $L$-Lipschitz, and uniformly $\varepsilon$-accurate on $C$. The construction uses only finitely many linear measurements $\langle b,\cdot\rangle_H$, with $b$ lying in a finite-dimensional subspace of $H$, and is exactly implementable by a $\operatorname{ReLU}$-MLP. Building on this, we introduce convex neural functionals (CNFs), a structured trainable architecture class containing our reconstruction, whose every admissible parameter configuration is automatically convex and Lipschitz, providing a principled foundation for learning convex functionals from finite data.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

The paper gives an explicit, finite-sample construction that turns point evaluations of a convex Lipschitz functional into a ReLU-MLP that stays exactly convex and L-Lipschitz with uniform error control.

read the letter

The main point is that for any compact convex set C in a separable Hilbert space, any L-Lipschitz convex functional on C, and any epsilon, the authors give a concrete way to build a convex L-Lipschitz approximant from finitely many exact point evaluations that is uniformly epsilon-close and exactly equal to a ReLU network. The construction takes a fine enough net on C, evaluates the functional there, then forms the pointwise supremum of all affine minorants with slope norm at most L that sit below those values. Compactness lets them reduce to a finite-dimensional subspace and a finite net on the dual ball, so only finitely many affines are needed and the whole thing becomes a max of linear forms, which a ReLU-MLP computes directly. This is new in the combination of exact preservation, finite samples, and neural-network realizability, and it supplies a clean theoretical basis for the convex neural functionals they introduce. The logic holds up without circularity or hidden assumptions; the error bound follows from the shared Lipschitz constant and the net density, and the finite-dimensional step uses standard compactness arguments in separable Hilbert spaces. The only real limitation is that everything assumes exact, noise-free evaluations, which is standard for this kind of existence result but means it does not directly address statistical estimation from noisy data. The paper is aimed at researchers who want structure-preserving approximations or convex-by-construction networks in optimization and machine learning. It is worth a serious referee because the claim is precise, the construction is explicit, and the supporting reasoning is internally consistent.

Referee Report

0 major / 3 minor

Summary. The manuscript claims that for every compact convex set C in a separable Hilbert space H, every L-Lipschitz convex functional ρ: C → ℝ, and every ε > 0, there exists an explicit reconstruction from finitely many linear measurements that is convex, exactly L-Lipschitz, and uniformly ε-accurate on C. The construction reduces to a finite-dimensional setting and is exactly realizable by a ReLU-MLP. The authors introduce Convex Neural Functionals (CNFs) as a trainable architecture class in which every admissible parameter set automatically yields a convex L-Lipschitz functional.

Significance. If the central construction holds, the result supplies a parameter-free, finite-sample procedure that preserves convexity and the precise Lipschitz constant while achieving arbitrary uniform accuracy in infinite-dimensional spaces. The explicit reduction to a maximum of finitely many affine forms (hence ReLU-MLP realizability) and the introduction of CNFs provide both theoretical guarantees and a structured inductive bias for learning convex functionals, with potential impact on risk measures, optimization, and structured machine learning.

minor comments (3)

The dependence of the sample size on the covering number of C, L, and ε is implicit in the δ-net argument but should be stated explicitly in the main theorem to clarify computational cost.
In the definition of the reconstruction as the pointwise supremum of affine minorants, verify that the finite net on the dual ball exactly preserves the Lipschitz constant L without relaxation.
The transition from the abstract reconstruction to the CNF architecture class would benefit from a precise statement of which parameter configurations correspond exactly to the finite-sample reconstructions.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive and accurate summary of our manuscript, as well as for highlighting its potential significance in providing structure-preserving approximations for convex Lipschitz functionals on Hilbert spaces. We appreciate the recommendation for minor revision. However, the report contains no specific major comments or points for clarification.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's central result is an explicit, direct construction: select a finite δ-net on the compact convex C with δ = ε/(2L), evaluate the given convex L-Lipschitz functional ρ at those points, and form the pointwise supremum of all affine minorants whose slopes have norm at most L and that lie below the observed values. This yields a convex L-Lipschitz function that is uniformly ε-close to ρ on C by the triangle inequality and net density. The finite-dimensional reduction and ReLU-MLP representation follow immediately because the resulting function is the maximum of finitely many affine forms. No parameter is fitted to data and then re-used as a prediction, no self-citation supplies a load-bearing uniqueness theorem, and no ansatz or renaming of a known result is invoked. The derivation is therefore self-contained and does not reduce to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 1 invented entities

The result rests on standard properties of separable Hilbert spaces and convexity/Lipschitz assumptions on the functional and domain; the main addition is the reconstruction construction and the CNF architecture class.

axioms (3)

standard math H is a separable Hilbert space
Domain for the functional and linear measurements.
domain assumption C is a compact convex subset of H
Required for uniform approximation on a bounded set.
domain assumption ρ is convex and L-Lipschitz on C
Properties that the reconstruction must preserve exactly.

invented entities (1)

Convex Neural Functionals (CNFs) no independent evidence
purpose: Trainable architecture class in which every admissible parameter set yields a convex and Lipschitz functional.
Introduced to contain the reconstruction and provide a foundation for learning convex functionals.

pith-pipeline@v0.9.0 · 5532 in / 1553 out tokens · 70908 ms · 2026-05-12T01:07:07.943609+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
fε,d(x) = max_m {⟨pm,x⟩ + min_n (ρ(ξn) - ⟨pm,ξn⟩)} is convex L-Lipschitz and ε-accurate (Thm 3.1); CNFs with nonnegative weights and max-pooling are automatically convex/Lipschitz (Thm 3.3)

Reference graph

Works this paper leans on

100 extracted references · 100 canonical work pages

[1]

arXiv preprint arXiv:2511.01125 , year=

One model to solve them all: 2BSDE families via neural operators , author=. arXiv preprint arXiv:2511.01125 , year=

work page arXiv
[2]

SIAM Journal on Financial Mathematics , volume=

Fully Dynamic Risk Measures: Horizon Risk, Time-Consistency, and Relations with BSDEs and BSVIEs , author=. SIAM Journal on Financial Mathematics , volume=. 2024 , publisher=

work page 2024
[3]

Acciaio, Beatrice and Kratsios, Anastasis and Pammer, Gudmund , TITLE =. Math. Finance , FJOURNAL =. 2024 , NUMBER =. doi:10.1111/mafi.12389 , URL =

work page doi:10.1111/mafi.12389 2024
[4]

arXiv preprint arXiv:2408.02853 , year=

SIG-BSDE for Dynamic Risk Measures , author=. arXiv preprint arXiv:2408.02853 , year=

work page arXiv
[5]

SIAM Journal on Financial Mathematics , volume=

Fully-dynamic risk-indifference pricing and no-good-deal bounds , author=. SIAM Journal on Financial Mathematics , volume=. 2020 , publisher=

work page 2020
[6]

Bernoulli , FJOURNAL =

Weed, Jonathan and Bach, Francis , TITLE =. Bernoulli , FJOURNAL =. 2019 , NUMBER =. doi:10.3150/18-BEJ1065 , URL =

work page doi:10.3150/18-bej1065 2019
[7]

2017 , PAGES =

Zhang, Jianfeng , TITLE =. 2017 , PAGES =. doi:10.1007/978-1-4939-7256-2 , URL =

work page doi:10.1007/978-1-4939-7256-2 2017
[8]

Boissard, Emmanuel and Le Gouic, Thibaut , TITLE =. Ann. Inst. Henri Poincar\'e. 2014 , NUMBER =. doi:10.1214/12-AIHP517 , URL =

work page doi:10.1214/12-aihp517 2014
[9]

Insurance: Mathematics and Economics , volume=

Risk measures via g-expectations , author=. Insurance: Mathematics and Economics , volume=. 2006 , publisher=

work page 2006
[10]

Pitman research notes in mathematics series , pages=

Backward SDE and related g-expectation , author=. Pitman research notes in mathematics series , pages=. 1997 , publisher=

work page 1997
[11]

Seminar on Stochastic Processes, 1985 , pages=

Brownian exit distribution of a ball , author=. Seminar on Stochastic Processes, 1985 , pages=. 1986 , organization=

work page 1985
[12]

arXiv preprint arXiv:2412.03405 , year=

Deep Operator BSDE: a Numerical Scheme to Approximate the Solution Operators , author=. arXiv preprint arXiv:2412.03405 , year=

work page arXiv
[13]

arXiv preprint arXiv:2407.18384 , year=

Mathematical theory of deep learning , author=. arXiv preprint arXiv:2407.18384 , year=

work page arXiv
[14]

Journal of Mathematical Analysis and Applications , volume=

An Extension Theorem for convex functions of class C1, 1 on Hilbert spaces , author=. Journal of Mathematical Analysis and Applications , volume=. 2017 , publisher=

work page 2017
[15]

Journal of Functional Analysis , volume=

Explicit formulas for C1, 1 and Cconv1, extensions of 1-jets in Hilbert and superreflexive spaces , author=. Journal of Functional Analysis , volume=. 2018 , publisher=

work page 2018
[16]

2019 , publisher=

Lipschitz functions , author=. 2019 , publisher=

work page 2019
[17]

1999 , publisher=

Convex Analysis and Variational Problems , author=. 1999 , publisher=

work page 1999
[18]

2004 , publisher=

Convex Optimization , author=. 2004 , publisher=

work page 2004
[19]

, author=

On general minimax theorems. , author=. Pacific J. Math. , volume=

work page
[20]

2017 , publisher =

Convex Analysis and Monotone Operator Theory in Hilbert Spaces , author =. 2017 , publisher =

work page 2017
[21]

Journal of the American Statistical Association , volume=

Point Estimates of Ordinates of Concave Functions , author=. Journal of the American Statistical Association , volume=. 1954 , doi=

work page 1954
[22]

The Annals of Statistics , volume=

Consistency in Concave Regression , author=. The Annals of Statistics , volume=. 1976 , doi=

work page 1976
[23]

The Annals of Statistics , volume=

Estimation of a Convex Function: Characterizations and Asymptotic Theory , author=. The Annals of Statistics , volume=. 2001 , doi=

work page 2001
[24]

The Annals of Statistics , volume=

Nonparametric Least Squares Estimation of a Multivariate Convex Regression Function , author=. The Annals of Statistics , volume=. 2011 , doi=

work page 2011
[25]

Journal of the American Statistical Association , volume=

A Computational Framework for Multivariate Convex Regression and Its Variants , author=. Journal of the American Statistical Association , volume=. 2019 , doi=

work page 2019
[26]

International Economic Review , volume=

The Construction of Utility Functions from Expenditure Data , author=. International Economic Review , volume=. 1967 , doi=

work page 1967
[27]

The Review of Economic Studies , volume=

Afriat and Revealed Preference Theory , author=. The Review of Economic Studies , volume=. 1973 , doi=

work page 1973
[28]

Econometrica , volume=

The Nonparametric Approach to Demand Analysis , author=. Econometrica , volume=. 1982 , doi=

work page 1982
[29]

Journal of Economic Theory , volume=

Afriat's Theorem for General Budget Sets , author=. Journal of Economic Theory , volume=. 2009 , doi=

work page 2009
[30]

The Journal of Business , volume=

Prices of State-Contingent Claims Implicit in Option Prices , author=. The Journal of Business , volume=. 1978 , doi=

work page 1978
[31]

Journal of Econometrics , volume=

Nonparametric Option Pricing under Shape Restrictions , author=. Journal of Econometrics , volume=. 2003 , doi=

work page 2003
[32]

Quantitative Finance , volume=

Arbitrage-Free Smoothing of the Implied Volatility Surface , author=. Quantitative Finance , volume=. 2009 , doi=

work page 2009
[33]

Arbitrage-Free

Gatheral, Jim and Jacquier, Antoine , journal=. Arbitrage-Free. 2014 , doi=

work page 2014
[34]

Mathematical Finance , volume=

Coherent Measures of Risk , author=. Mathematical Finance , volume=. 1999 , doi=

work page 1999
[35]

Finance and Stochastics , volume=

Convex Measures of Risk and Trading Constraints , author=. Finance and Stochastics , volume=. 2002 , doi=

work page 2002
[36]

Journal of Banking & Finance , volume=

Putting Order in Risk Measures , author=. Journal of Banking & Finance , volume=. 2002 , doi=

work page 2002
[37]

Risk Measures via

Rosazza Gianin, Emanuela , journal=. Risk Measures via. 2006 , doi=

work page 2006
[38]

Mathematische Annalen , volume=

A General Version of the Fundamental Theorem of Asset Pricing , author=. Mathematische Annalen , volume=. 1994 , doi=

work page 1994
[39]

SIAM Journal on Control and Optimization , volume=

Dynamic Programming and Pricing of Contingent Claims in an Incomplete Market , author=. SIAM Journal on Control and Optimization , volume=. 1995 , doi=

work page 1995
[40]

Probability Theory and Related Fields , volume=

Optional Decomposition of Supermartingales and Hedging Contingent Claims in Incomplete Security Markets , author=. Probability Theory and Related Fields , volume=. 1996 , doi=

work page 1996
[41]

Stochastic Finance: An Introduction in Discrete Time , author=

work page
[42]

arXiv preprint arXiv:2601.11209 , year=

SANOS--Smooth strictly Arbitrage-free Non-parametric Option Surfaces , author=. arXiv preprint arXiv:2601.11209 , year=

work page arXiv
[43]

The Journal of FinTech , volume=

On deep calibration of (rough) stochastic volatility models , author=. The Journal of FinTech , volume=. 2025 , publisher=

work page 2025
[44]

Journal of Machine Learning Research , volume=

Multivariate Convex Regression with Adaptive Partitioning , author=. Journal of Machine Learning Research , volume=

work page
[45]

Mathematics of Control, Signals and Systems , volume=

Approximation by Superpositions of a Sigmoidal Function , author=. Mathematics of Control, Signals and Systems , volume=. 1989 , doi=

work page 1989
[46]

Neural Networks , volume=

Approximation Capabilities of Multilayer Feedforward Networks , author=. Neural Networks , volume=. 1991 , doi=

work page 1991
[47]

Neural Networks , volume=

Multilayer Feedforward Networks with a Nonpolynomial Activation Function Can Approximate Any Function , author=. Neural Networks , volume=. 1993 , doi=

work page 1993
[48]

Acta Numerica , volume=

Approximation Theory of the MLP Model in Neural Networks , author=. Acta Numerica , volume=. 1999 , doi=

work page 1999
[49]

IEEE Transactions on Information Theory , volume=

Universal Approximation Bounds for Superpositions of a Sigmoidal Function , author=. IEEE Transactions on Information Theory , volume=. 1993 , doi=

work page 1993
[50]

Neural Networks , volume=

Error Bounds for Approximations with Deep ReLU Networks , author=. Neural Networks , volume=. 2017 , doi=

work page 2017
[51]

Neural Networks , volume=

Optimal Approximation of Piecewise Smooth Functions Using Deep ReLU Neural Networks , author=. Neural Networks , volume=. 2018 , doi=

work page 2018
[52]

Error Bounds for Approximations with Deep ReLU Neural Networks in W^

G. Error Bounds for Approximations with Deep ReLU Neural Networks in W^. Analysis and Applications , volume=. 2020 , doi=

work page 2020
[53]

Analysis and Applications , volume=

Deep ReLU Networks and High-Order Finite Element Methods , author=. Analysis and Applications , volume=. 2020 , doi=

work page 2020
[54]

Acta Numerica , volume=

Neural Network Approximation , author=. Acta Numerica , volume=. 2021 , doi=

work page 2021
[55]

Proceedings of the 29th Annual Conference on Learning Theory , series=

Benefits of Depth in Neural Networks , author=. Proceedings of the 29th Annual Conference on Learning Theory , series=

work page
[56]

Advances in Neural Information Processing Systems , volume=

The Expressive Power of Neural Networks: A View from the Width , author=. Advances in Neural Information Processing Systems , volume=

work page
[57]

2017 , eprint=

Approximating Continuous Functions by ReLU Nets of Minimal Width , author=. 2017 , eprint=

work page 2017
[58]

Proceedings of Thirty Third Conference on Learning Theory , series=

Universal Approximation with Deep Narrow Networks , author=. Proceedings of Thirty Third Conference on Learning Theory , series=

work page
[59]

The Annals of Statistics , volume=

Nonparametric Regression Using Deep Neural Networks with ReLU Activation Function , author=. The Annals of Statistics , volume=. 2020 , doi=

work page 2020
[60]

IEEE Transactions on Neural Networks , volume=

Universal Approximation to Nonlinear Operators by Neural Networks with Arbitrary Activation Functions and Its Application to Dynamical Systems , author=. IEEE Transactions on Neural Networks , volume=. 1995 , doi=

work page 1995
[61]

arXiv preprint arXiv:2412.05109 , year=

Generating rectifiable measures through neural networks , author=. arXiv preprint arXiv:2412.05109 , year=

work page arXiv
[62]

Annals of Mathematics and Artificial Intelligence , volume=

The Universal Approximation Property , author=. Annals of Mathematics and Artificial Intelligence , volume=. 2021 , doi=

work page 2021
[63]

Advances in Neural Information Processing Systems , volume=

Non-Euclidean Universal Approximation , author=. Advances in Neural Information Processing Systems , volume=

work page
[64]

Journal of Machine Learning Research , volume=

Universal Approximation Theorems for Differentiable Geometric Deep Learning , author=. Journal of Machine Learning Research , volume=

work page
[65]

International Conference on Learning Representations , year=

Universal Approximation Under Constraints is Possible with Transformers , author=. International Conference on Learning Representations , year=

work page
[66]

Neurocomputing , volume=

Learning Sub-Patterns in Piecewise Continuous Functions , author=. Neurocomputing , volume=. 2022 , doi=

work page 2022
[67]

Transactions on Machine Learning Research , year=

Do ReLU Networks Have An Edge When Approximating Compactly-Supported Functions? , author=. Transactions on Machine Learning Research , year=

work page
[68]

arXiv preprint arXiv:2409.12335 , year=

Bridging the Gap Between Approximation and Learning via Optimal Approximation by ReLU MLPs of Maximal Regularity , author=. arXiv preprint arXiv:2409.12335 , year=

work page arXiv
[69]

Journal of Machine Learning Research , volume=

NEU: A Meta-Algorithm for Universal UAP-Invariant Feature Representation , author=. Journal of Machine Learning Research , volume=

work page
[70]

Journal of Machine Learning Research , volume=

Instance-Dependent Generalization Bounds via Optimal Transport , author=. Journal of Machine Learning Research , volume=

work page
[71]

Optimization and Engineering , volume=

Convex Piecewise-Linear Fitting , author=. Optimization and Engineering , volume=. 2009 , doi=

work page 2009
[72]

Proceedings of the Eighteenth International Conference on Artificial Intelligence and Statistics , series=

Near-Optimal Max-Affine Estimators for Convex Regression , author=. Proceedings of the Eighteenth International Conference on Artificial Intelligence and Statistics , series=. 2015 , publisher=

work page 2015
[73]

SIAM Journal on Mathematics of Data Science , year=

Max-Affine Regression via First-Order Methods , author=. SIAM Journal on Mathematics of Data Science , year=

work page
[74]

Whitney Extension Theorems for Convex Functions of the Classes

Azagra, Daniel and Mudarra, Carlos , journal=. Whitney Extension Theorems for Convex Functions of the Classes

work page
[75]

The Sharp Whitney Extension Theorem for Convex

Mudarra, Carlos , year=. The Sharp Whitney Extension Theorem for Convex. 2602.05642 , archivePrefix=

work page arXiv
[76]

Proceedings of the 34th International Conference on Machine Learning , series =

Input Convex Neural Networks , author =. Proceedings of the 34th International Conference on Machine Learning , series =. 2017 , publisher =

work page 2017
[77]

Advances in Neural Information Processing Systems , volume =

Monotonic Networks , author =. Advances in Neural Information Processing Systems , volume =. 1997 , publisher =

work page 1997
[78]

International Conference on Learning Representations , year =

Optimal Control Via Neural Networks: A Convex Approach , author =. International Conference on Learning Representations , year =

work page
[79]

Advances in Neural Information Processing Systems , volume =

Certified Monotonic Neural Networks , author =. Advances in Neural Information Processing Systems , volume =

work page
[80]

Proceedings of the 30th International Conference on Machine Learning , series =

Maxout Networks , author =. Proceedings of the 30th International Conference on Machine Learning , series =. 2013 , publisher =

work page 2013

Showing first 80 references.