arXiv subjects

Maxime Larcher

Publications and source records attributed to Maxime Larcher.

8 recordsLinked to original sources

arXiv2023

Hardest Monotone Functions for Evolutionary Algorithms

In this paper we revisit the question how hard it can be for the $(1+1)$ Evolutionary Algorithm to optimize monotone pseudo-Boolean functions. By introducing a more pessimistic stochastic process, the partially-ordered evolutionary algorithm (PO-EA) model, Jansen first proved a runtime bound of $O(n^{3/2})$. More recently, Lengler, Martinsson and Steger improved this upper bound to $O(n \log^2 n)$ by an entropy compression argument. In this work, we analyze monotone functions that may adversarially vary at each step of the optimization, so-called dynamic monotone functions. We introduce the function Switching Dynamic BinVal (SDBV) and prove, using a combinatorial argument, that for the $(1+1)$-EA with any mutation rate $p \in [0,1]$, SDBV is drift minimizing within the class of dynamic monotone functions. We further show that the $(1+1)$-EA optimizes SDBV in $\Theta(n^{3/2})$ generations. Therefore, our construction provides the first explicit example which realizes the pessimism of the \poea model. Our simulations demonstrate matching runtimes for both the static and the self-adjusting $(1,\lambda)$-EA and $(1+\lambda)$-EA. Moreover, devising an example for fixed dimension, we illustrate that drift minimization does not equal maximal runtime beyond asymptotic analysis.

cs.NE↗

arXiv2023

Gated recurrent neural networks discover attention

Recent architectural developments have enabled recurrent neural networks (RNNs) to reach and even surpass the performance of Transformers on certain sequence modeling tasks. These modern RNNs feature a prominent design pattern: linear recurrent layers interconnected by feedforward paths with multiplicative gating. Here, we show how RNNs equipped with these two design elements can exactly implement (linear) self-attention, the main building block of Transformers. By reverse-engineering a set of trained RNNs, we find that gradient descent in practice discovers our construction. In particular, we examine RNNs trained to solve simple in-context learning tasks on which Transformers are known to excel and find that gradient descent instills in our RNNs the same attention-based in-context learning algorithm used by Transformers. Our findings highlight the importance of multiplicative interactions in neural networks and suggest that certain RNNs might be unexpectedly implementing attention under the hood.

cs.LG↗

arXiv2022

OneMax is not the Easiest Function for Fitness Improvements

We study the $(1:s+1)$ success rule for controlling the population size of the $(1,\lambda)$-EA. It was shown by Hevia Fajardo and Sudholt that this parameter control mechanism can run into problems for large $s$ if the fitness landscape is too easy. They conjectured that this problem is worst for the OneMax benchmark, since in some well-established sense OneMax is known to be the easiest fitness landscape. In this paper we disprove this conjecture and show that OneMax is not the easiest fitness landscape with respect to finding improving steps. As a consequence, we show that there exists $s$ and $\varepsilon$ such that the self-adjusting $(1,\lambda)$-EA with $(1:s+1)$-rule optimizes OneMax efficiently when started with $\varepsilon n$ zero-bits, but does not find the optimum in polynomial time on Dynamic BinVal. Hence, we show that there are landscapes where the problem of the $(1:s+1)$-rule for controlling the population size of the $(1, \lambda)$-EA is more severe than for OneMax.

cs.NE↗

arXiv2022

Self-adjusting Population Sizes for the $(1, \lambda)$-EA on Monotone Functions

We study the $(1,\lambda)$-EA with mutation rate $c/n$ for $c\le 1$, where the population size is adaptively controlled with the $(1:s+1)$-success rule. Recently, Hevia Fajardo and Sudholt have shown that this setup with $c=1$ is efficient on \onemax for $s<1$, but inefficient if $s \ge 18$. Surprisingly, the hardest part is not close to the optimum, but rather at linear distance. We show that this behavior is not specific to \onemax. If $s$ is small, then the algorithm is efficient on all monotone functions, and if $s$ is large, then it needs superpolynomial time on all monotone functions. In the former case, for $c<1$ we show a $O(n)$ upper bound for the number of generations and $O(n\log n)$ for the number of function evaluations, and for $c=1$ we show $O(n\log n)$ generations and $O(n^2\log\log n)$ evaluations. We also show formally that optimization is always fast, regardless of $s$, if the algorithm starts in proximity of the optimum. All results also hold in a dynamic environment where the fitness function changes in each generation.

cs.NE↗

arXiv2022

Solving Static Permutation Mastermind using $O(n \log n)$ Queries

Permutation Mastermind is a version of the classical mastermind game in which the number of positions $n$ is equal to the number of colors $k$, and repetition of colors is not allowed, neither in the codeword nor in the queries. In this paper we solve the main open question from Glazik, Jäger, Schiemann and Srivastav (2021), who asked whether their bound of $O(n^{1.525})$ for the static version can be improved to $O(n \log n)$, which would be best possible. By using a simple probabilistic argument we show that this is indeed the case.

math.CO↗

arXiv2021

Note on Long Paths in Eulerian Digraphs

Long Paths and Cycles in eulerian digraphs have gotten a lot of attention recently. In this short note, we show how to use methods from Knierim, Larcher, Martinsson, Noever (2021) to find paths of length $d/(\log d+1)$ in Eulerian digraphs with average degree $d$, improving the recent result of $Ω(d^{1/2+1/40})$. Our result is optimal up to at most a logarithmic factor.

math.CO↗

arXiv2020

Maker-Breaker Games on Random Hypergraphs

In this paper, we study Maker-Breaker games on the random hypergraph $H_{n,s,p}$, obtained from the complete $s$-graph by keeping every edge independently with probability $p$. We determine the threshold probability for the property of Maker winning the game as a function of $s$, the uniformity of the underlying hypergraph, as well as $m$, $b$, the number of vertices that Maker and Breaker are respectively allowed to pick each turn. In addition, we show that depending on those $m,b,s$, there are two types of thresholds: either being Maker-win is a local property and the threshold is weak, or it is related to global properties of the random hypergraph and the threshold is semi-sharp. We conjecture that in the latter case, the threshold is actually sharp.

math.CO↗

arXiv2020

Long Cycles, Heavy Cycles and Cycle Decompositions in Digraphs

Hajós conjectured in 1968 that every Eulerian $n$-vertex graph can be decomposed into at most $\lfloor (n-1)/2\rfloor$ edge-disjoint cycles. This has been confirmed for some special graph classes, but the general case remains open. In a sequence of papers by Bienia and Meyniel (1986), Dean (1986), and Bollobás and Scott (1996) it was analogously conjectured that every \emph{directed} Eulerian graph can be decomposed into $O(n)$ cycles. In this paper, we show that every directed Eulerian graph can be decomposed into $O(n \log Δ)$ disjoint cycles, thus making progress towards the conjecture by Bollobás and Scott. Our approach is based on finding heavy cycles in certain edge-weightings of directed graphs. As a further consequence of our techniques, we prove that for every edge-weighted digraph in which every vertex has out-weight at least $1$, there exists a cycle with weight at least $Ω(\log \log n/{\log n})$, thus resolving a question by Bollobás and Scott.

math.CO↗