arXiv subjects

Maximilian Merkert

Publications and source records attributed to Maximilian Merkert.

3 recordsLinked to original sources

arXiv2026

Complexity of partitioned-items response problems: matchings and perfect matchings

We consider bilevel optimization problems in which leader and follower jointly construct a feasible solution for an underlying combinatorial optimization problem. Response problems ask whether the leader can encourage -- or, in the pessimistic setting, enforce -- a reaction of the follower that includes a set of mandatory items while excluding a set of forbidden items. Our investigation focuses on tractability results for various cases which emerge from different combinations of the total number of mandatory, forbidden, and neutral items. After providing some results for response problems that hold for any underlying combinatorial optimization problem, we examine response problems over the maximum-weight matching problem and the minimum-weight perfect matching problem as illustrative and surprisingly varied examples. Among other results, we show that the response problem is hard for even a single given mandatory or forbidden edge. On the other hand, it is fixed-parameter tractable with respect to the total number of non-mandatory edges. If, however, each follower's edge is either mandatory or forbidden, the response problem for the perfect matching problem is solvable in polynomial time while it remains NP-hard for the maximum-weight matching problem.

math.OC↗

arXiv2025

On the Expressiveness of Rational ReLU Neural Networks With Bounded Depth

To confirm that the expressive power of ReLU neural networks grows with their depth, the function $F_n = \max \{0,x_1,\ldots,x_n\}$ has been considered in the literature. A conjecture by Hertrich, Basu, Di Summa, and Skutella [NeurIPS 2021] states that any ReLU network that exactly represents $F_n$ has at least $\lceil\log_2 (n+1)\rceil$ hidden layers. The conjecture has recently been confirmed for networks with integer weights by Haase, Hertrich, and Loho [ICLR 2023]. We follow up on this line of research and show that, within ReLU networks whose weights are decimal fractions, $F_n$ can only be represented by networks with at least $\lceil\log_3 (n+1)\rceil$ hidden layers. Moreover, if all weights are $N$-ary fractions, then $F_n$ can only be represented by networks with at least $Ω( \frac{\ln n}{\ln \ln N})$ layers. These results are a partial confirmation of the above conjecture for rational ReLU networks, and provide the first non-constant lower bound on the depth of practically relevant ReLU networks.

cs.LG↗

arXiv2024

Binary Cyclic Transversal Polytopes

With every family of finitely many subsets of a finite-dimensional vector space over the Galois-field with two elements we associate a cyclic transversal polytope. It turns out that those polytopes generalize several well-known polytopes that are relevant in combinatorial optimization, among them cut polytopes as well as stable set and matching polytopes. We introduce the class of lifted odd-set inequalities and prove results demonstrating their strength. In particular, we show that they suffice to describe cyclic transversal polytopes if the union of the sets in the family has rank at most two. We also describe extended formulations for cyclic transversal polytopes and introduce a special relaxation hierarchy for them.

math.CO↗