Frank-Wolfe Recursions for the Emergency Response Problem on Measure Spaces

Di Yu; Raghu Pasupathy; Shane G. Henderson

REVIEW 4 major objections 6 minor 2 cited by

Frank-Wolfe Recursions for the Emergency Response Problem on Measure Spaces

T0 review · 4 major / 6 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 a fully-corrective Frank-Wolfe algorithm on the infinite-dimensional space of measures solves the OHCA volunteer-allocation problem, with $O(1/\sqrt{k})$ decay of influence values at selected locations even when…

desk verdict The paper's influence function drops the x∈S condition from the stated objective, so the central claim and the numerical evidence for it do not hold; the fc-FW convergence analysis is still a useful standalone contribution. read the letter →

arxiv 2507.09808 v1 pith:5LGHB7TF submitted 2025-07-13 math.OC stat.CO

Di Yu , Shane G. Henderson , Raghu Pasupathy This is my paper

classification math.OCstat.CO MSC 90C4890C25

keywords measureoptimizationFrank-Wolferecursionfully-correctivealgorithminfluencefunctionvonMisesderivativeout-of-hospitalcardiacarrestemergencyresponseP-means

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper considers the problem of choosing a spatial probability distribution of volunteers over a city to minimize the probability of death from out-of-hospital cardiac arrest. It claims that this optimum can be found by a fully-corrective Frank-Wolfe algorithm that works directly in the infinite-dimensional space of measures, without first discretizing the city or imposing a parametric family. The central theoretical result is that even when the per-iteration task of minimizing the influence function—the location-wise first-order score—is solved only approximately, the influence values at the selected locations tend to zero at rate $O(1/\sqrt{k})$, and if solved globally the rate improves to $O(1/k)$. The paper also proves the objective is convex, the feasible set is compact, an optimum exists, and that under the $L_1$ travel metric with discrete demand the optimal measure is supported on a finite grid. If these claims are right, volunteer recruitment targets for cities can be computed directly on continuous maps, and the method transfers to the broader class of $P$-means problems.

What carries the argument

The machinery is the influence function together with the fully-corrective Frank-Wolfe recursion it drives. For a measure $\mu$, the influence function $h_\mu(x)$ is the location-wise rate at which $J$ changes when a small point mass is moved to $x$, and Eq. (18) gives its closed form. The fc-FW algorithm keeps a growing set of scaled Dirac atoms, adds an atom at a point $x^*(\mu_k)$ with $h_{\mu_k}(x^*(\mu_k))\le 0$, and then solves the finite-dimensional convex problem of minimizing $J$ over the convex hull of all accumulated atoms. Lemma 1 shows that $L$-smoothness plus the non-positive-influence condition yields a sufficient decrease of the objective at every step, and Theorem 2 turns this into influence values decaying as $O(1/\sqrt{k})$; when the influence subproblem is solved globally, the objective value converges at $O(1/k)$. For the $L_1$ metric, Lemma 3 shows the influence function is strictly concave on each rectangle of the demand-coordinate grid, which forces its minima, and hence the optimal support, onto the grid vertices.

What would settle it

Take the three-vertex equilateral example of Section 5.2 and compute the directional derivative at the centroid $o$ by finite differences, $(J(\mu+t(b\delta_o-\mu))-J(\mu))/t$ for small $t$, comparing it with Eq. (18); if the two disagree for points outside the demand set, the influence function used by the algorithm is not the derivative the paper derives, and the descent and convergence guarantees as stated do not govern those iterates.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the infinite-dimensional emergency-response problem (P), minimize $J(\mu)=\int_S \eta(dy)\int_0^\infty \exp(-\mu(B(y,t)\cap S))\,d\beta(t)$ over nonnegative Borel measures with total mass $b$, is tractable by adaptive particle accumulation rather than a priori discretization. The paper proves that $J$ is convex and $L$-smooth in total variation, that the feasible set is weakly compact, and that optimal measures exist; the first-order object is the influence function $h_\mu(x)$, which Theorem 7 identifies as the von Mises derivative of $J$. The algorithmic claim is that a fully-corrective Frank-Wolfe recursion—add the atom $b\delta_x$ at a point where $h_{\mu_k}(x)\le 0$, then re-optimize the weights over all accumulated atoms—produces a convergent sequence: $h_{\mu_k}(x^*_k)\to 0$ at rate $O(1/\sqrt{k})$, with no need to globally minimize the nonconvex influence function. The numerical portion claims that even simple three- or four-point instances yield non-obvious optimal measures, that continuous and Auckland-scale cases run in hundreds to thousands of iterations, and that with the $L_1$ metric the influence function is piecewise strictly concave, so the optimal support lies on the grid built from demand coordinates.

Load-bearing premise

The load-bearing premise is that the influence function $h_\mu(x)$ in Eq. (18) is the true first-order derivative of the death-probability objective for every candidate volunteer location in the convex hull of the incident region; the derivation counts a trial mass at $x$ toward every ball $B(y,t)$ when $\|y-x\|\le t$, even when $x$ lies outside the incident region, where that mass cannot intersect $B(y,t)\cap S$.

Editorial extensions

If this is right

City-scale volunteer allocation can be solved directly on continuous maps, so optimality no longer depends on a chosen area-unit discretization.
At each iteration any point with non-positive influence suffices; a global solve of the nonconvex influence subproblem is not needed for the $O(1/\sqrt{k})$ guarantee.
If the influence subproblem is solved globally, the objective-value convergence improves to $O(1/k)$, recovering the standard Frank-Wolfe rate in the infinite-dimensional setting.
With $L_1$ travel and discrete incident locations, the optimal volunteer measure is concentrated on the $O(n^2)$ grid vertices determined by demand coordinates, enabling lookup-style algorithms.
The framework transfers to $P$-means problems, whose objectives share the same exponential-of-negative-measure form.

Reading between the lines

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

An extension the paper leaves implicit: for candidate volunteer locations outside the incident region, the factor counting whether $x$ lies in $B(y,t)\cap S$ should be zero; rerunning the three- and four-point experiments with that corrected derivative would test whether the reported off-demand-point support persists.
The proof machinery is not tied to cardiac arrest: any convex, $L$-smooth objective on measures with a computable influence function should admit the same fully-corrective Frank-Wolfe treatment, making the algorithm a candidate for other spatial resource-allocation and coverage problems.
The $L_1$ grid-support theorem suggests a practical speedup the paper does not test: restrict each influence-function search to the $O(n^2)$ grid vertices and benchmark it against the continuous Adam-based searches on the Auckland data.

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

The central claim of this paper doesn't hold up. The influence function in Theorem 7, Eq. (18), drops the x∈S condition that appears in \bar B(y,t)=B(y,t)∩S. For a perturbation bδ_x, the correct first-order term is b·1{x∈S, ||y-x||≤t} - µ(\bar B(y,t)); Eq. (18) writes b·1_{[||y-x||,∞)}(t), omitting x∈S. For any x in cvx(S)\S, the influence function is too large by a nonnegative amount. The algorithm selects new support points by minimizing h_µ, so points outside S look spuriously attractive. Section 5.2's negative influence at the centroid is exactly this artifact: the centroid lies outside S, so the correct influence is positive. The claimed conclusion that the optimal measure is not supported on the demand points therefore doesn't follow from problem (P). The L1 finite-support theorem inherits the same mismatch.

What's genuinely new and good: the fc-FW convergence analysis (Lemma 1, Theorem 2) is abstract and, as far as I can tell, correct. It shows a sufficient decrease from h(x*)≤0 alone, with an O(1/√k) rate on the influence values. That's a useful contribution independent of the specific application. The restriction-to-cvx(S) argument (Theorem 3) and the convexity/existence results are standard but fine, though the existence proof should verify lower semicontinuity carefully. The L1 piecewise-concavity observation is interesting, and the Auckland experiments suggest the method scales; but no code or data are included, and there is no baseline comparison.

Minor issue: Theorem 2 claims ∑|h|<∞, but the proof gives only ∑h²<∞. That's an overclaim, though it doesn't affect the main rate.

This is a repairable paper. The fix is either to remove the ∩S from the objective (so volunteers everywhere count) or to add the x∈S indicator in the influence function. But as submitted, the derivation breaks the central claim, and the numerical results in Sections 5.2 and 6 are not evidence for problem (P).

If I were the editor, I'd send it to review because the fc-FW theorem deserves airing, and the authors may well fix the derivative. But I'd expect major revision at best. For a reading group, it's a good case study in how a dropped indicator can sink an infinite-dimensional optimization paper; otherwise it's not something I'd cite yet.

Referee Report

4 major / 6 minor

Summary. The paper treats the problem of allocating a fixed total mass b of volunteers, modeled as the intensity measure of a spatial Poisson process, to minimize the expected probability of death from out-of-hospital cardiac arrest (OHCA). The authors formulate an infinite-dimensional convex optimization problem over measures supported on the convex hull of the incident region, derive an influence function as the first-order variational object, and analyze a fully-corrective Frank-Wolfe (fc-FW) algorithm that iteratively adds Dirac atoms. They prove structural properties (compactness of the feasible set, convexity of the objective, existence of optima), establish a sufficient-decrease lemma and a sublinear convergence rate under a weak subproblem condition, present numerical experiments on synthetic and Auckland data, and prove an L1-norm support-reduction result. The central claim is that the method solves the infinite-dimensional problem without discretization and that optimal measures have nontrivial support even for a few demand points.

Significance. If the results were correct, the paper would offer a useful extension of Frank-Wolfe methods to measure-valued optimization in an applied setting, with the attractive feature that the subproblem needs only a point of non-positive influence rather than a global minimizer. The city-scale numerical study and the L1-norm structural result would support practical relevance. However, the core variational object, the influence function, is incorrect for the stated problem, and the main analytic and numerical conclusions rely on that object. Consequently, the convergence results describe an algorithm that minimizes a different functional than the one in problem (P), and the claimed advances are not established for the stated problem.

major comments (4)

[Theorem 7, Eq. (18)] Equation (18) defines h_µ(x) with the term -b I_{[∥y-x∥,∞)}(t) for every x∈cvx(S). Since \bar B(y,t)=B(y,t)∩S in (2) and (P), the correct first-order term under the perturbation bδ_x is -b 1{x∈S} I_{[∥y-x∥,∞)}(t); for x∈cvx(S)\S, the atom δ_x contributes zero to every µ(\bar B(y,t)). Thus (18) is valid only on S, and the von Mises derivative identity (19)/(23) fails for ν with mass outside S. Because Algorithm 1 minimizes h_µk over all of cvx(S), the subproblem in Step 3 is not the variational problem for (P); consequently Lemma 1 and Theorem 2 concern a different algorithm, and the claimed O(1/√k) convergence is not a statement about the stated problem.
[§5.2, Eq. (28)] Section 5.2 evaluates the influence at the centroid o∈cvx(S)\S using the incorrect Eq. (18). The correct influence at o is ∫_S η(dy)∫_0^∞ \tilde µ(\bar B(y,t)) exp(-\tilde µ(\bar B(y,t))) dβ(t) > 0, since the -b indicator term is absent. The negative value in (28) is an artifact of the missing x∈S condition. Moreover, for the objective (2), any mass placed at x∉S contributes to no \bar B(y,t); moving it to a point of S with positive λ_i strictly decreases J, so an optimal measure for (P) is supported on S. The conclusion that supp(µ*) is not contained in {y1,y2,y3} is therefore not just unproved but incompatible with the stated model.
[§7, Lemma 3 and Theorem 10] Section 7 repeats the same error: the influence function (34) is only valid for x∈S. For x in the interior of a rectangle Sj,k that does not meet the finite demand set S, the correct h_µ(x) is constant in x. Therefore Lemma 3's strict concavity inequality (36) is false (both sides are equal), Lemma 4's claim that minimizers lie at rectangle vertices is false (a constant is minimized everywhere), and Theorem 10's proof collapses. The support statement for discrete η may be true for the trivial reason that mass off S is wasted, but the derivation via strict concavity and the claimed vertex look-up / support-reduction procedure are not valid for problem (P).
[§6, Algorithm 1 implementation] All numerical experiments in Section 6 use the influence function in the form (29), which is derived from the incorrect (18). The reported optimal measures with mass in the interior of cvx(S) cannot be optimal for (P), since such mass is absent from every \bar B(y,t); the contour plots of h in Figures 2(b) and 2(d) certify optimality with respect to the wrong functional. The implementation therefore does not solve the problem introduced in (P), and the claims of 'complex solution structure' and 'scalability' are not supported by the experiments.

minor comments (6)

[Theorem 6 proof] The proof says 'J is convex by Theorem 3'; the correct reference for convexity is Theorem 5.
[§3.2, item (b)] The text says 'Theorem 5 guarantees that h_µk(x*) → 0'; the intended reference is Theorem 2.
[Theorem 10 proof] The phrase 'Theorem 4 directly implies' should read 'Lemma 4'.
[Definition 3] The notation δ_x := IA(x) is ambiguous; it should be the set function δ_x(A) = 1{x∈A}.
[§5.2] The text refers to 'Figure 2' before any figure is introduced; the intended reference appears to be Figure 1, the triangle diagram.
[§3.2] The paper relies on [33] for Theorem 1, Theorem 8, Lemma 2, and Lemma 6 but does not clearly identify these as imported results; a remark at the start of Section 3.2 would remove ambiguity.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; the main flagged problem (Eq. 18 dropping x∈S) is a non-circular derivation error, and the same-author citations are to standard, independently checkable lemmas.

full rationale

The paper's central fc-FW result is genuinely new: Lemma 1 and Theorem 2 derive a sufficient decrease and O(1/sqrt(k)) decay of h_μ(x*) from L-smoothness without requiring global subproblem optimality; the proof is in the paper and does not define the target quantity in terms of itself. The L-smoothness inequality used in Lemma 1 is quoted from the authors' [33], but it is a parameter-free calculus lemma whose stated assumptions (sup-norm Lipschitz influence functions) do not include the convergence claim, so under the review rules this citation is independent support rather than circularity. The same holds for the optimality condition (Theorem 8) and support lemma (Lemma 2) imported from [33]: they are standard first-order KKT-type characterizations for convex measure optimization, not uniqueness theorems imported to forbid alternatives. No fitted parameter is relabeled as a prediction: Section 6.4 explicitly states 'our results should not be construed as providing realistic predictions' and uses published incident estimates 'without re-estimation.' The reviewer's flagged flaw is real but not circular: setting ν=bδ_x in the paper's own (21) gives (µ−ν)(\bar B(y,t)) = µ(\bar B(y,t)) − b·1{||y−x||≤t}·1{x∈S}, whereas Eq. (18) omits the factor 1{x∈S}; hence h_μ is not the von Mises derivative of the stated objective on cvx(S)\S, and Section 5.2's negative centroid influence and the claimed L1 support theorem do not follow from (P) as written. This is an algebraic correctness/consistency failure, not a reduction by construction, so it does not raise the circularity score beyond a minor allowance for the load-bearing same-author citations.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

Central theoretical and numerical claims rest on the Poisson void probability model, the concavity of beta, and the correctness of the influence function. The latter is inconsistent with the stated objective (see red flags). The free parameters are empirical inputs borrowed from prior literature, not fitted in this paper. No new entities are introduced.

free parameters (2)

beta(t) coefficients (0.679, 0.262) = 0.679, 0.262
Death probability function from De Maio et al. [8], used in all experiments; parameters fit to survival data in prior literature, not by this paper.
Auckland per-area incident rates lambda_l = published estimates from [29]
Used for the Auckland case study; estimated from GoodSAM 2013-2020 data by van den Berg et al. [29], not re-estimated.

assumptions (5)

domain assumption beta is continuous, increasing, strictly concave on [0, infinity)
Section 1.1; used in Theorem 9 and Lemma 3. The adopted logistic death function is concave for t >= 0.
domain assumption Volunteer locations follow a spatial Poisson point process with intensity measure mu
Section 1.1; gives the void probability exp(-mu(B)) used in (2).
domain assumption S is compact and eta is a Borel probability measure on S
Section 1.1; needed for weak* compactness of M+(cvx(S), b).
standard math Descent inequality for L-smooth functionals (Lemma 6 from [33])
Lemma 6 is cited from the authors' prior work [33]; L-smoothness is verified in Appendix B with L = 2b + 1.
standard math Dominated convergence and Riesz representation justify the derivative exchange and weak* compactness
Used in Theorem 7 and Appendix A.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Frank-Wolfe Recursions for the Emergency Response Problem on Measure Spaces." pith.science (2026). https://pith.science/paper/5LGHB7TF

@misc{pith2026250709808,
  author       = {Pith},
  title        = {Pith review of: Frank-Wolfe Recursions for the Emergency Response Problem on Measure Spaces},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5LGHB7TF}},
  note         = {Machine review of arXiv:2507.09808}
}

abstract

We consider an optimization problem over measures for emergency response to out-of-hospital cardiac arrest (OHCA), where the goal is to allocate volunteer resources across a spatial region to minimize the probability of death. The problem is infinite-dimensional and poses challenges for analysis and computation. We first establish structural properties, including convexity of the objective functional, compactness of the feasible set, and existence of optimal solutions. We also derive the influence function, which serves as the first-order variational object in our optimization framework. We then adapt and analyze a fully-corrective Frank-Wolfe (fc-FW) algorithm that operates directly on the infinite-dimensional problem without discretization or parametric approximation. We show a form of convergence even when subproblems are not solved to global optimality. Our full implementation of fc-FW demonstrates complex solution structure even in simple discrete cases, reveals nontrivial volunteer allocations in continuous cases, and scales to realistic urban scenarios using OHCA data from the city of Auckland, New Zealand. Finally, we show that when volunteer travel is modeled through the $L_1$ norm, the influence function is piecewise strictly concave, enabling fast computation via support reduction. The proposed framework and analysis extend naturally to a broad class of $P$-means problems.

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

An Occupation-Measure and Frank-Wolfe Framework for Heterogeneous Mean-Field Control
math.OC 2026-07 conditional novelty 5.0 of 10

Heterogeneous mean-field control is convex under a matrix-valued PSD kernel condition and can be solved by a Frank–Wolfe scheme that decomposes into parallel classical optimal-control problems.
Convexifying Mean-Field Control: An Occupation-Measure and Frank-Wolfe Approach
math.OC 2026-07 conditional novelty 5.0 of 10

A Frank-Wolfe algorithm on occupation measures solves mean-field control as a convex program whose iterations reduce to classical optimal control subproblems.

Reference graph

Works this paper leans on

33 extracted references · 30 canonical work pages · cited by 2 Pith papers

[1]

Bækgaard JS, Viereck S, Møller TP, Ersbøll AK, Lippert F, Folke F (2017) The effects of public access defib- rillation on survival after out-of-hospital cardiac arrest: a systematic review of observational studies.Circulation 136(10):954–965

work page 2017
[2]

SIAM Journal on Optimization 27(2)

Boyd N, Schiebinger G, Recht B (2017) The alternating descent conditional gradient method for sparse inverse problems. SIAM Journal on Optimization 27(2)

work page 2017
[3]

Bredies K, Pikkarainen H (2013) Inverse problems in spaces of measures. ESAIM. Control, optimisation and calculus of variations 19(1):190–218, ISSN 1292-8119

work page 2013
[4]

(2015) Convex optimization: Algorithms and complexity

Bubeck S, et al. (2015) Convex optimization: Algorithms and complexity. Foundations and Trends® in Machine Learning 8(3-4):231–357

work page 2015
[5]

Mathematical pro- gramming 194(1-2):487–532, ISSN 0025-5610

Chizat L (2022) Sparse optimization on measures with over-parameterized gradient descent. Mathematical pro- gramming 194(1-2):487–532, ISSN 0025-5610

work page 2022
[6]

Advances in neural information processing systems 31

Chizat L, Bach F (2018) On the global convergence of gradient descent for over-parameterized models using optimal transport. Advances in neural information processing systems 31

work page 2018
[7]

Chu C, Blanchet J, Glynn P (2019) Probability functional descent: A unifying perspective on GANs, variational inference, and reinforcement learning

work page 2019
[8]

(2003) Optimal defibrillation response intervals for maximum out-of-hospital cardiac arrest survival rates

De Maio VJ, Stiell IG, Wells GA, Spaite DW, Group OPALSS, et al. (2003) Optimal defibrillation response intervals for maximum out-of-hospital cardiac arrest survival rates. Annals of emergency medicine 42(2):242– 250

work page 2003

Show all 33 references

[9]

Inverse Problems 36(1):014001

Denoyelle Q, Duval V , Peyr ´e G, Soubies E (2019) The sliding Frank–Wolfe algorithm and its application to super-resolution microscopy. Inverse Problems 36(1):014001

2019
[10]

Journal of Math- ematical Analysis and Applications 62(2):432–444

Dunn JC, Harshbarger S (1978) Conditional gradient algorithms with open loop step size rules. Journal of Math- ematical Analysis and Applications 62(2):432–444

1978
[11]

Eftekhari A, Thompson A (2019) Sparse inverse problems over measures: Equivalence of the conditional gradient and exchange methods

2019
[12]

Von Mises Calculus for Statistical Functionals 19

Fernholz L (1983) Lecture notes in statistics. Von Mises Calculus for Statistical Functionals 19

1983
[13]

Fernholz L (2011) Functional derivatives in statistics: Asymptotics and robustness

2011
[14]

https://www.goodsamapp.org/, accessed February 29, 2020

GoodSAM Platform (2020) GoodSAM. https://www.goodsamapp.org/, accessed February 29, 2020

2020
[15]

Kent C, Blanchet J, Glynn P (2021) Frank-Wolfe methods in probability space. 30

2021
[16]

Pro- ceedings of the fourth Berkeley symposium on mathematical statistics and probability, volume 1, 381–405 (Univ of California Press)

Kiefer J (1960) Optimum experimental designs V, with applications to systematic and rotatable designs. Pro- ceedings of the fourth Berkeley symposium on mathematical statistics and probability, volume 1, 381–405 (Univ of California Press)

1960
[17]

The Annals of Statistics 2(5):849–879

Kiefer J (1974) General equivalence theory for optimum designs (approximate theory). The Annals of Statistics 2(5):849–879

1974
[18]

URL https://arxiv.org/abs/ 1412.6980

Kingma DP, Ba J (2017) Adam: A method for stochastic optimization. URL https://arxiv.org/abs/ 1412.6980

2017 arXiv
[19]

Proceedings of the National Academy of Sciences 115(33), URL http://dx.doi.org/10.1073/pnas

Mei S, Montanari A, Nguyen PM (2018) A mean field view of the landscape of two-layer neural networks. Proceedings of the National Academy of Sciences 115(33), URL http://dx.doi.org/10.1073/pnas. 1806579115

2018 doi
[20]

Statistics and Computing 12:115–123

Molchanov I, Zuyev S (2002) Steepest descent algorithms in a space of measures. Statistics and Computing 12:115–123

2002
[21]

Nichol G, Stiell IG, Laupacis A, De Maio VJ, Wells GA, et al. (1999) A cumulative meta-analysis of the effec- tiveness of defibrillator-capable emergency medical services for victims of out-of-hospital cardiac arrest.Annals of emergency medicine 34(4):517–525

1999
[22]

arXiv preprint arXiv:1712.05438

Nitanda A, Suzuki T (2017) Stochastic particle gradient descent for infinite ensembles. arXiv preprint arXiv:1712.05438

2017 arXiv
[23]

Wiley series in probability and statistics

Okabe A (2000) Spatial tessellations: concepts and applications of Voronoi diagrams. Wiley series in probability and statistics. Applied probability and statistics section (Chichester: Wiley), 2nd edition, ISBN 0471986356

2000
[24]

(2019) First-response treatment after out-of-hospital cardiac arrest: a survey of current prac- tices across 29 countries in Europe

Oving I, Masterson S, Tjelmeland IB, Jonsson M, Semeraro F, Ringh M, Truhlar A, Cimpoesu D, Folke F, Beesems SG, et al. (2019) First-response treatment after out-of-hospital cardiac arrest: a survey of current prac- tices across 29 countries in Europe. Scandinavian Journal of ...

2019
[25]

https://www.pulsepoint.org/, accessed: 2020-02-29

PulsePoint (2020) Pulsepoint. https://www.pulsepoint.org/, accessed: 2020-02-29

2020
[26]

Resnick S (1987) Extreme Values, Regular Variation, and Point Processes(New York, NY: Springer)

1987
[27]

Resnick S (1992) Adventures in Stochastic Processes (Boston, MA.: Birkh¨auser)

1992
[28]

Circulation: Cardiovascular Quality and Outcomes 3(1):63–81

Sasson C, Rogers MA, Dahl J, Kellermann AL (2010) Predictors of survival from out-of-hospital cardiac arrest: a systematic review and meta-analysis. Circulation: Cardiovascular Quality and Outcomes 3(1):63–81

2010
[29]

Management Science

van den Berg PL, Henderson SG, Jagtenberg CJ, Li H (2024) Modeling the impact of community first responders. Management Science

2024
[30]

Resuscitation 51(2):113–122

Waalewijn RA, de V os R, Tijssen JG, Koster RW (2001) Survival models for out-of-hospital cardiopulmonary resuscitation from the perspectives of the bystander, the first responder, and the paramedic. Resuscitation 51(2):113–122. 31

2001
[31]

Critical care 24:1–13

Yan S, Gan Y , Jiang N, Wang R, Chen Y , Luo Z, Zong Q, Chen S, Lv C (2020) The global survival rate among adult out-of-hospital cardiac arrest patients who received cardiopulmonary resuscitation: a systematic review and meta-analysis. Critical care 24:1–13

2020
[32]

The Annals of Statistics 52(4):1774–1795

Yan Y , Wang K, Rigollet P (2024) Learning Gaussian mixtures using the Wasserstein–Fisher–Rao gradient flow. The Annals of Statistics 52(4):1774–1795

2024
[33]

URL https://arxiv.org/abs/2407.00307

Yu D, Henderson SG, Pasupathy R (2024) Deterministic and stochastic Frank-Wolfe recursion on probability spaces. URL https://arxiv.org/abs/2407.00307. 32

2024

Pith tools

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

[1] [1]

Bækgaard JS, Viereck S, Møller TP, Ersbøll AK, Lippert F, Folke F (2017) The effects of public access defib- rillation on survival after out-of-hospital cardiac arrest: a systematic review of observational studies.Circulation 136(10):954–965

work page 2017

[2] [2]

SIAM Journal on Optimization 27(2)

Boyd N, Schiebinger G, Recht B (2017) The alternating descent conditional gradient method for sparse inverse problems. SIAM Journal on Optimization 27(2)

work page 2017

[3] [3]

Bredies K, Pikkarainen H (2013) Inverse problems in spaces of measures. ESAIM. Control, optimisation and calculus of variations 19(1):190–218, ISSN 1292-8119

work page 2013

[4] [4]

(2015) Convex optimization: Algorithms and complexity

Bubeck S, et al. (2015) Convex optimization: Algorithms and complexity. Foundations and Trends® in Machine Learning 8(3-4):231–357

work page 2015

[5] [5]

Mathematical pro- gramming 194(1-2):487–532, ISSN 0025-5610

Chizat L (2022) Sparse optimization on measures with over-parameterized gradient descent. Mathematical pro- gramming 194(1-2):487–532, ISSN 0025-5610

work page 2022

[6] [6]

Advances in neural information processing systems 31

Chizat L, Bach F (2018) On the global convergence of gradient descent for over-parameterized models using optimal transport. Advances in neural information processing systems 31

work page 2018

[7] [7]

Chu C, Blanchet J, Glynn P (2019) Probability functional descent: A unifying perspective on GANs, variational inference, and reinforcement learning

work page 2019

[8] [8]

(2003) Optimal defibrillation response intervals for maximum out-of-hospital cardiac arrest survival rates

De Maio VJ, Stiell IG, Wells GA, Spaite DW, Group OPALSS, et al. (2003) Optimal defibrillation response intervals for maximum out-of-hospital cardiac arrest survival rates. Annals of emergency medicine 42(2):242– 250

work page 2003