An Efficient Evolutionary Algorithm for Minimum Cost Submodular Cover

Victoria G. Crawford

REVIEW 2 major objections 4 minor 23 references

An Efficient Evolutionary Algorithm for Minimum Cost Submodular Cover

T0 review · 2 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read An evolutionary algorithm achieves the first polynomial-time constant bicriteria guarantee for Minimum Cost Submodular Cover.

desk verdict Solid, first-of-its-kind EA guarantee for submodular cover; theory holds up, but the bin-index definition needs an integer repair before it's publishable. read the letter →

arxiv 1908.01029 v1 pith:C4XUKVG2 submitted 2019-08-02 cs.DS

Victoria G. Crawford This is my paper · ORCID

classification cs.DS

keywords minimumcostsubmodularcoverevolutionaryalgorithmbicriteriaapproximationexpectedpolynomialtimesubmodularitycost-effectivenessinfluencethresholdproblem

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 introduces EASC, an evolutionary algorithm for Minimum Cost Submodular Cover (MCSC), and proves that it finds a near-feasible, near-optimal solution in expected polynomial time. For any $\epsilon \in (0,1)$, EASC returns a set $A$ with benefit $f(A) \ge (1-\epsilon)\tau$ and cost $c(A) \le (\ln(1/\epsilon)+1)c(A^*)$, where $A^*$ is an optimal cover. This is the first such polynomial-time guarantee for an evolutionary algorithm on MCSC; earlier evolutionary approaches either had exponential worst-case time or required the benefit function to take only polynomially many distinct values. The algorithm partitions the benefit range into logarithmically spaced bins and keeps one cost-effective solution per bin, with a comparison rule modeled on greedy submodular cover. Experiments on influence-threshold instances show EASC improving on the greedy solution and converging faster than the existing Pareto optimization method on several networks.

What carries the argument

The machinery rests on a bin function that cuts $[0,\tau]$ into $\log_\delta(\epsilon)+1$ intervals of the form $[(1-\delta^i)\tau,(1-\delta^{i+1})\tau)$, with the final bin holding all sets with $f(X) \ge (1-\epsilon)\tau$; each bin holds at most one solution, so the population stays at $O((c_{\max}/c_{\min})\ln(1/\epsilon)n)$. Within a bin, the comparison uses the cost-effectiveness $\varphi(X)=c(X)/\ln(\tau/(\tau-f(X)))$ (with $\varphi(X)=c(X)$ in the first and last bins). This $\varphi$ is chosen so that a solution with $\varphi(X)\le c(A^*)$ is as cost-effective as any greedy step; the bin spacing matches the greedy marginal-gain lemma, so adding the best marginal element always advances a cost-effective solution into a strictly higher bin. Two submodularity lemmas carry the argument: $\tau - f_\tau(X\cup\{x^*\}) \le (1 - c(x^*)/c(A^*))(\tau - f_\tau(X))$, and $c(x^*) \le c(A^*)$.

What would settle it

A counterexample would be an MCSC instance with known optimum $A^*$, $\epsilon$ constant, and $\delta$ chosen in the permitted range $[1 - c_{\min}/c(A^*),\, 1 - c_{\min}/c(S)]$, for which EASC's expected number of iterations to first place a solution in the final bin exceeds $e n ((c_{\max}/c_{\min}) \ln(1/\epsilon) n + 1)^2$ by more than a constant factor; such instances could be produced by exhaustive search over small ground sets and tested by repeated simulation.

Watch

Extended reading notes

Core claim

The central result is Theorem 1: for an MCSC instance with optimal solution $A^* \ne \emptyset$, if EASC is run with $\epsilon \in (0,1)$ and $\delta \in [1 - c_{\min}/c(A^*),\, 1 - c_{\min}/c(S)]$, then within an expected $e n ((c_{\max}/c_{\min}) \ln(1/\epsilon) n + 1)^2$ iterations its population contains a set $A$ in the final bin with $f(A) \ge (1-\epsilon)\tau$ and $c(A) \le (\ln(1/\epsilon)+1)c(A^*)$. The proof shows that the highest bin containing a cost-effective set never decreases, and that a specific mutation -- adding the element with maximum marginal benefit per unit cost -- advances that bin by at least one step. When $c_{\max}/c_{\min}$ is polynomial in $n$ and $\epsilon$ is constant, the expected time is polynomial, which is the first such guarantee for an evolutionary algorithm on MCSC.

Load-bearing premise

The guarantee depends on the user supplying a bin parameter $\delta$ that is at least $1 - c_{\min}/c(A^*)$, and since $c(A^*)$ is unknown this requires an upper bound $B$ on the optimum cost with $B \le c(S)$, typically obtained from a greedy solution; if no such bound is available, the proof's bin-progress step fails.

Editorial extensions

If this is right

When $c_{\max}/c_{\min}$ is polynomially bounded in $n$ and $\epsilon$ is constant, EASC reaches a solution meeting the guarantee in expected polynomial time; the population size stays at $O((c_{\max}/c_{\min})\ln(1/\epsilon)n)$ throughout.
Once a qualifying solution enters the final bin, the comparison rule in that bin uses only cost, so the guarantee is permanent.
Because the same $(1-\epsilon,\ \ln(1/\epsilon)+1)$ bicriteria guarantee holds for the greedy algorithm, EASC matches the worst-case guarantee while experiments show it can improve on the greedy solution.
The bin structure and cost-effectiveness comparison are designed from monotone submodularity and are presented as generalizable ideas for other submodular optimization problems.

Reading between the lines

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

If the bin-progress argument generalizes to other monotone submodular objectives (for example submodular maximization under matroid or knapsack constraints), the same logarithmic bin schedule could yield polynomial-time evolutionary guarantees for those problems, replacing generic Pareto populations.
The iteration bound scales with $(c_{\max}/c_{\min})^2$; for instances with exponentially skewed element costs the guarantee stops being polynomial, so the practical scope is limited to cost profiles within a polynomial factor of each other.
A tighter or looser upper bound $B$ on $c(A^*)$ changes $\delta$ and therefore the number of bins; the theory only needs $B \le c(S)$, so testing convergence against noisy or loose bounds (for example $B=2c(\text{greedy})$) would probe how robust the practical runtime is to the parameter choice.

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.

Desk Editor's Note

read the letter

Colleague,

Punchline: this is a real result—the first evolutionary algorithm for MCSC with an expected polynomial-time, constant bicriteria guarantee—and the theory holds up. But the paper has one definitional gap that must be fixed, and the experiments are too thin to carry much weight. Still, it deserves a serious referee, not a desk rejection.

What's new: the binning scheme plus the φ cost-effectiveness measure are original, and the analysis showing a bounded population size suffices is a genuine improvement over POM, whose expected time depends on N, the number of distinct f values (possibly exponential). The author credits the prior binning ideas and the greedy analysis properly. The proof of Theorem 1 is carefully structured, and the lemmas are standard but correctly proven. The parameter δ can be set from any upper bound on c(A*) that is ≤ c(S), which is a practical workaround given c(A*) is unknown.

The main soft spot is the integer-bin issue. Section 2.1 defines the final bin as log_δ(ε), but for arbitrary δ, ε ∈ (0,1), log_δ(ε) is not an integer. As written, bin(X) is not defined for every X—there are gaps between the last non-final interval and the final threshold. This is not just a proof typo; Algorithm 1 calls bin(X') and would be undefined on some inputs. The repair is straightforward: let r be an integer (e.g., r = ceil(log_δ ε)), make the final bin [1−δ^r, 1], and adjust the proof accordingly. Every step of the proof appears to survive; the final threshold becomes stricter than (1−ε)τ, which only helps feasibility, and the expected-time bound still holds up to constants. So it's a fixable flaw, but it must be fixed before publication.

The experiments average only three runs per instance, with no error bars or significance tests. The claim of faster convergence than POM on some datasets is plausible but unproven. That's a minor-to-moderate weakness.

The reader's take is mostly right; the stress-test note about the bin index is correct and worth taking seriously.

Bottom line: this paper is for the evolutionary computation and submodular optimization communities. I'd bring it to my reading group and would cite it if I worked on EA guarantees, though I'd want the bin definition cleaned up first. It deserves peer review, with the referee asked to confirm the integer-bin fix and to check that the δ-satisfiability condition is stated cleanly.

Referee Report

2 major / 4 minor

Summary. The paper proposes EASC, an evolutionary algorithm for the Minimum Cost Submodular Cover (MCSC) problem. EASC maintains a population with at most one solution per benefit bin, compares solutions in the same bin by a cost-effectiveness measure φ, and mutates a uniformly random population member at each step. The main theoretical contribution is Theorem 1, which claims that, when run indefinitely with ε∈(0,1) and δ in a specific interval, EASC eventually contains a set A with f(A)≥(1−ε)τ and c(A)≤(ln(1/ε)+1)c(A*) in expected O(n^3 ((cmax/cmin) ln(1/ε))^2) iterations. The proof is organized in five parts, establishing monotonicity of a cost-effectiveness progress measure, bin progress via the greedy marginal-gain lemma, preservation of cost-effectiveness, and an expected-iteration bound. The paper also gives an experimental comparison with the greedy algorithm and the POM evolutionary algorithm on four influence-threshold instances. The exposition is generally clear, and the appendix contains proofs of the standard submodularity lemmas. However, the central algorithm and theorem are not well-defined for generic δ and ε because the bin structure uses the generally non-integer quantity log_δ(ε) as both a bin count and a final bin index; this needs a repair. A second, smaller gap appears in one subcase of the proof of Theorem 1. Both issues are fixable without changing the claimed asymptotic result.

Significance. If Theorem 1 holds after the necessary repairs, the paper makes a significant contribution: EASC would be the first evolutionary algorithm with an expected polynomial-time constant bicriteria approximation guarantee for MCSC, avoiding the polynomial-population assumption required by POM. The proof strategy, especially the cost-effectiveness comparison and the use of bin progress, is a genuinely useful idea that may extend to other submodular problems. The paper is also honest in noting in Proposition 2 that the same bicriteria guarantee already holds for the greedy algorithm, so the theoretical novelty is the algorithmic mechanism rather than a new approximation factor. The public code and the careful inclusion of the standard lemmas in the appendix are strengths. The experimental section is illustrative rather than decisive, since it averages only three runs and reports no variance or statistical tests.

major comments (2)

[Section 2.1, Algorithm 1, Theorem 1] The bin function and the algorithm are not well-defined for generic input parameters. In the definition of bin(X), the index set {0,...,log_δ(ε)−1} and the final case 'bin(X)=log_δ(ε)' require r=log_δ(ε) to be an integer, but for arbitrary δ∈(0,1) and ε∈(0,1) this is generally false; for example, δ=0.5 and ε=0.1 give r≈3.32. As a result, many subsets X are not assigned to any bin, and Algorithm 1's call to bin(X') is undefined. The proof in Section 3 also silently assumes that r is an integer and that bin a+1 exists for every a<r. The manuscript needs a repair, for instance by defining q=ceil(log_δ(ε)), indexing bins 0,...,q, giving the final bin priority whenever f(X)≥(1−ε)τ, and adjusting the population-size bound from q+1 to q+2. With such a repair, the proof steps in Parts One through Five appear to go through, but the current text does not state the repair.
[Section 3, Part Three, Case (i)] The proof that {x*} is cost-effective when bin({x*})<r is incomplete in the subcase bin({x*})=0. The displayed inequality τ−fτ(x*)≤exp(−c(x*)/c(A*))τ can be rearranged to φ({x*})≤c(A*) only when φ is the ratio c/ln(τ/(τ−f)), but for bin 0 the definition in Section 2.2 sets φ({x*})=c({x*}) instead. The missing argument is supplied by Lemma 2, which gives c(x*)≤c(A*) because f(∅)<τ; invoking Lemma 2 would complete the subcase. As written, this subcase is not proved, although the repair is straightforward.

minor comments (4)

[Section 2, penultimate paragraph] The sentence 'Lower ϵ and δ values result in less bins' is backwards for ε: decreasing ε increases ln(1/ε) and hence increases the number of bins. Only decreasing δ reduces the bin count.
[Section 4] The experimental results average only three runs per algorithm and report no error bars, variance, or significance tests. Claims such as 'EASC converges faster than POM' in Section 4.2 should be softened or quantitatively supported.
[Figure 1] The figure labels the final bin as 'logδ(ϵ)', which is the same non-integer quantity that causes the well-definedness problem in Section 2.1; the figure should be updated to match the repaired integer bin count.
[Theorem 1] The stated expected-iteration bound of en((cmax/cmin)ln(1/ε)n + 1)^2 depends on the bin-count bound. If the repair uses q=ceil(log_δ(ε)), the additive constant becomes 2, so the bound should be restated to be consistent with the repaired definition.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: EASC's guarantee is proven from independent submodularity lemmas; the δ parameter is a standard conditional scaling choice, not a fitted prediction.

full rationale

EASC's Theorem 1 is not circular. The proof in Section 3 defines cost-effective sets and shows via Lemmas 1 and 2, which are proven in Appendix 5.2 from monotone submodularity, that cost-effective solutions are never replaced by non-cost-effective ones, that a lower-bin solution can mutate into a strictly higher bin with polynomially bounded probability, and that a final-bin solution satisfies c(A) ≤ (ln(1/ε)+1)c(A*). The f(A) ≥ (1−ε)τ part is true by definition of the final bin, but the cost bound is derived, not assumed. The input δ is required to lie in an interval depending on c(A*), but the paper explicitly shows how to instantiate it from an upper bound B on c(A*) (e.g., the cost of the greedy solution), which is a standard parameter-setting device and does not feed the target solution into the algorithm or the proof. No parameter is fitted to experimental outcomes, and the lemmas are standard and self-contained. The non-integer bin-index issue for generic δ and ε (r = log_δ(ε) need not be an integer) is a genuine well-definedness and correctness defect in the statement, but it is not a circularity: it does not make any claimed result an input to itself.

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

No data-fitted free parameters appear in the central result. The algorithm's parameters are epsilon (standard approximation input) and delta, which is set via an upper bound on the optimal cost, a practical requirement rather than a fit to data. All axioms are standard submodularity facts or explicit instance conditions. No new entities are introduced.

free parameters (2)

delta = 1 - cmin/B, where B is an upper bound on c(A*) with B <= c(S)
Controls bin widths; the proof's progress argument (Part Two) requires delta >= 1 - cmin/c(A*). It is not fitted to data, but it is a hand-chosen parameter tied to the unknown optimal solution cost.
epsilon = input parameter in (0,1)
Standard approximation slack; not fitted, but determines the trade-off between feasibility and cost.

assumptions (5)

domain assumption f is monotone submodular
Definition of MCSC instance; used throughout.
standard math f_tau = min(f, tau) is monotone submodular
Truncation preserves monotonicity and submodularity; used in Lemmas 1 and 2.
standard math There exists an a* in A* with Delta f_tau(X,a*)/c(a*) >= Delta f_tau(X,A*)/c(A*)
Proven in Lemma 2 via submodularity; used to bound the cost of the element added in mutation.
domain assumption delta >= 1 - cmin/c(A*)
Input condition that makes each bin's width at least cmin/c(A*) of the remaining gap; without it Part Two fails.
domain assumption A* is not empty
Non-trivial cover; required for c(A*) in denominators.

how reviews work

0 comments

Cite this review

Pith. "Pith review of An Efficient Evolutionary Algorithm for Minimum Cost Submodular Cover." pith.science (2026). https://pith.science/paper/C4XUKVG2

@misc{pith2026190801029,
  author       = {Pith},
  title        = {Pith review of: An Efficient Evolutionary Algorithm for Minimum Cost Submodular Cover},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C4XUKVG2}},
  note         = {Machine review of arXiv:1908.01029}
}

read the original abstract

In this paper, the Minimum Cost Submodular Cover problem is studied, which is to minimize a modular cost function such that the monotone submodular benefit function is above a threshold. For this problem, an evolutionary algorithm EASC is introduced that achieves a constant, bicriteria approximation in expected polynomial time; this is the first polynomial-time evolutionary approximation algorithm for Minimum Cost Submodular Cover. To achieve this running time, ideas motivated by submodularity and monotonicity are incorporated into the evolutionary process, which likely will extend to other submodular optimization problems. In a practical application, EASC is demonstrated to outperform the greedy algorithm and converge faster than competing evolutionary algorithms for this problem.

Figures

Figures reproduced from arXiv: 1908.01029 by the authors.

**Figure 1.** The region [0, τ ] is discretized into logδ () + 1 bins in EASC. Solutions X ⊆ S are mapped into the bin corresponding to the region where min{f(X), τ} falls. B contains at most 1 subset of S for each bin. 2 Evolutionary Algorithm for MCSC (EASC) In this section, the algorithm EASC (Evolutionary Algorithm for Submodular Cover) is introduced. Pseudocode for EASC can be found in Algorithm 1. EASC is designed for find… view at source ↗

**Figure 2.** An illustration of the cost-effectiveness [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Over the duration of EASC and POM, the minimum cost of any solution in the population with [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 22 canonical work pages

[1]

Badanidiyuru and J

A. Badanidiyuru and J. Vondr \'a k. Fast algorithms for maximizing submodular functions. In Proceedings of the twenty-fifth annual ACM-SIAM symposium on Discrete algorithms, pages 1497--1514. SIAM, 2014

work page 2014
[2]

Borgs, M

C. Borgs, M. Brautbar, J. Chayes, and B. Lucier. Maximizing social influence in nearly optimal time. In Proceedings of the twenty-fifth annual ACM-SIAM symposium on Discrete algorithms, pages 946--957. SIAM, 2014

work page 2014
[3]

W. Chen, C. Wang, and Y. Wang. Scalable influence maximization for prevalent viral marketing in large-scale social networks. In Proceedings of the 16th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 1029--1038. ACM, 2010

work page 2010
[4]

U. Feige. A threshold of ln n for approximating set cover. Journal of the ACM (JACM), 45 0 (4): 0 634--652, 1998

1998
[5]

Friedrich and F

T. Friedrich and F. Neumann. Maximizing submodular functions under matroid constraints by multi-objective evolutionary algorithms. In International Conference on Parallel Problem Solving from Nature, pages 922--931. Springer, 2014

work page 2014
[6]

Friedrich and F

T. Friedrich and F. Neumann. Maximizing submodular functions under matroid constraints by evolutionary algorithms. Evolutionary computation, 23 0 (4): 0 543--558, 2015

work page 2015
[7]

Friedrich, A

T. Friedrich, A. G \"o bel, F. Quinzan, and M. Wagner. Heavy-tailed mutation operators in single-objective combinatorial optimization. In International Conference on Parallel Problem Solving from Nature, pages 134--145. Springer, 2018

work page 2018
[8]

Goyal, F

A. Goyal, F. Bonchi, L. V. Lakshmanan, and S. Venkatasubramanian. On minimizing budget and time in influence propagation over social networks. Social network analysis and mining, 3 0 (2): 0 179--192, 2013

work page 2013

Show all 23 references

[9]

Guillory and J

A. Guillory and J. A. Bilmes. Simultaneous learning and covering with adversarial noise. In ICML, volume 11, pages 369--376, 2011

2011
[10]

Horoba and F

C. Horoba and F. Neumann. Additive approximations of pareto-optimal sets by evolutionary multi-objective algorithms. In Proceedings of the tenth ACM SIGEVO workshop on Foundations of genetic algorithms, pages 79--86. ACM, 2009

2009
[11]

R. K. Iyer and J. A. Bilmes. Submodular optimization with submodular cover and submodular knapsack constraints. In Advances in Neural Information Processing Systems, pages 2436--2444, 2013

2013
[12]

Kempe, J

D. Kempe, J. Kleinberg, and \'E . Tardos. Maximizing the spread of influence through a social network. In Proceedings of the ninth ACM SIGKDD international conference on Knowledge discovery and data mining, pages 137--146. ACM, 2003

2003
[13]

Kuhnle, T

A. Kuhnle, T. Pan, M. A. Alim, and M. T. Thai. Scalable bicriteria algorithms for the threshold activation problem in online social networks. In INFOCOM 2017-IEEE Conference on Computer Communications, IEEE, pages 1--9. IEEE, 2017

2017
[14]

Laumanns, L

M. Laumanns, L. Thiele, K. Deb, and E. Zitzler. Combining convergence and diversity in evolutionary multiobjective optimization. Evolutionary computation, 10 0 (3): 0 263--282, 2002

2002
[15]

Leskovec and A

J. Leskovec and A. Krevl. \ SNAP Datasets \ : \ Stanford \ large network dataset collection. 2015

2015
[16]

Mirzasoleiman, A

B. Mirzasoleiman, A. Karbasi, A. Badanidiyuru, and A. Krause. Distributed submodular cover: Succinctly summarizing massive data. In Advances in Neural Information Processing Systems, pages 2881--2889, 2015

2015
[17]

Mirzasoleiman, M

B. Mirzasoleiman, M. Zadimoghaddam, and A. Karbasi. Fast distributed submodular cover: Public-private data summarization. In Advances in Neural Information Processing Systems, pages 3594--3602, 2016

2016
[18]

G. L. Nemhauser, L. A. Wolsey, and M. L. Fisher. An analysis of approximations for maximizing submodular set functions—i. Mathematical programming, 14 0 (1): 0 265--294, 1978

1978
[19]

Norouzi-Fard, A

A. Norouzi-Fard, A. Bazzi, I. Bogunovic, M. El Halabi, Y.-P. Hsieh, and V. Cevher. An efficient streaming algorithm for the submodular cover problem. In Advances in Neural Information Processing Systems, pages 4493--4501, 2016

2016
[20]

C. Qian, Y. Yu, and Z.-H. Zhou. On constrained boolean pareto optimization. In IJCAI, pages 389--395, 2015 a

2015
[21]

C. Qian, Y. Yu, and Z.-H. Zhou. Subset selection by pareto optimization. In Advances in Neural Information Processing Systems, pages 1774--1782, 2015 b

2015
[22]

Qian, J.-C

C. Qian, J.-C. Shi, Y. Yu, and K. Tang. On subset selection with general cost constraints. In Proceedings of the 26th International Joint Conference on Artificial Intelligence (IJCAI-2017), pages 2613--2619, 2017

2017
[23]

L. A. Wolsey. An analysis of the greedy algorithm for the submodular set covering problem. Combinatorica, 2 0 (4): 0 385--393, 1982

1982

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Badanidiyuru and J

A. Badanidiyuru and J. Vondr \'a k. Fast algorithms for maximizing submodular functions. In Proceedings of the twenty-fifth annual ACM-SIAM symposium on Discrete algorithms, pages 1497--1514. SIAM, 2014

work page 2014

[2] [2]

Borgs, M

C. Borgs, M. Brautbar, J. Chayes, and B. Lucier. Maximizing social influence in nearly optimal time. In Proceedings of the twenty-fifth annual ACM-SIAM symposium on Discrete algorithms, pages 946--957. SIAM, 2014

work page 2014

[3] [3]

W. Chen, C. Wang, and Y. Wang. Scalable influence maximization for prevalent viral marketing in large-scale social networks. In Proceedings of the 16th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 1029--1038. ACM, 2010

work page 2010

[4] [4]

U. Feige. A threshold of ln n for approximating set cover. Journal of the ACM (JACM), 45 0 (4): 0 634--652, 1998

1998

[5] [5]

Friedrich and F

T. Friedrich and F. Neumann. Maximizing submodular functions under matroid constraints by multi-objective evolutionary algorithms. In International Conference on Parallel Problem Solving from Nature, pages 922--931. Springer, 2014

work page 2014

[6] [6]

Friedrich and F

T. Friedrich and F. Neumann. Maximizing submodular functions under matroid constraints by evolutionary algorithms. Evolutionary computation, 23 0 (4): 0 543--558, 2015

work page 2015

[7] [7]

Friedrich, A

T. Friedrich, A. G \"o bel, F. Quinzan, and M. Wagner. Heavy-tailed mutation operators in single-objective combinatorial optimization. In International Conference on Parallel Problem Solving from Nature, pages 134--145. Springer, 2018

work page 2018

[8] [8]

Goyal, F

A. Goyal, F. Bonchi, L. V. Lakshmanan, and S. Venkatasubramanian. On minimizing budget and time in influence propagation over social networks. Social network analysis and mining, 3 0 (2): 0 179--192, 2013

work page 2013