The Cellular Direct Search is introduced, a derivative-free optimization method inspired by the emergent dynamics of Cellular Automata that establishes CDS as a viable and interpretable paradigm for closed-box optimization tasks where gradient information is unavailable or unreliable.
Abstract
We introduce the Cellular Direct Search (CDS), a derivative-free optimization method inspired by the emergent dynamics of Cellular Automata. CDS partitions the parameter space into a discrete grid where a population of agents evolves according to strictly local rules: a Jealous Neighbor Rule for competitive selection and a Solitary Regeneration Rule for adaptive exploration. From a theoretical perspective, we formally ground CDS in the theory of Directional Direct Search, proving that its local neighborhood constitutes a maximal positive basis. Under Lipschitz smoothness assumptions, we establish a formal guarantee of $O(\|\mathbf {h}\|)$ -approximate first-order stationarity for the interior points reached by the algorithm on a fixed grid of resolution h. Moreover, we interpret the imposed hyperspherical search boundary through the lens of Lagrangian duality, providing an intuitive link to $L_{2}$ -type regularization under Slater’s condition, and derive a heuristic proportionality between grid resolution and gradient-descent learning rates. We evaluate CDS against a comprehensive suite of modern and classical baselines across standardized benchmark problems from BBOB and the complete ABS and Layeb benchmark families, including shifted and rotated instances, as well as on a real-world convolutional neural network hyperparameter optimization task. The experimental results characterize the strengths and limitations of CDS across diverse optimization landscapes. Additional robustness analyses indicate that the observed behavior cannot be explained solely by center-location or coordinate-alignment effects. Furthermore, CDS demonstrates robust behavior in high-dimensional settings, maintaining stable performance even under constrained evaluation budgets. This establishes CDS as a viable and interpretable paradigm for closed-box optimization tasks where gradient information is unavailable or unreliable.
We study obstacle-constrained variational problems whose reduced energies depend on a probability law through a lower-level optimizer. Uniform strong convexity yields a single-valued Lipschitz follower response, while convexity and a Poincare inequality give a unique upper-level policy. A type-Lipschitz reduced margina...
Dynamical optimal mass transportation over nonlinear systems is often hindered by the curse of dimensionality and open-loop implementation. We propose an alternative framework that sacrifices exact optimality in favor of stability, robustness, and scalability. We use a stochastic port-Hamiltonian system as a reference...
We study a mean-field planning problem for a swarm of interacting, non-cooperative agents whose dynamics are stochastic and kinetic. The noise acts on the velocity only, and the interaction (of Cucker-Smale type for instance) enters through the drift. For a fixed flow, the problem reduces to a Schr\"odinger bridge prob...
How search strategies evolve in finite, depletable landscapes remains a question in foraging theory. We study this problem with an evolutionary simulation in which agents forage on a two-dimensional toroidal lattice containing non-renewable resources distributed uniformly or as L\'evy dust. Each agent carries a heritab...
Shailendra Bhandari, Alexander Szorkovszky, Anis Yazidi et al.· 0 citations
We introduce a general class of interacting reinforced stochastic systems evolving on a network. The dynamics are governed by two distinct interaction structures: a sampling network, which determines how information is acquired, and a reinforcement network, which determines how the sampled information influences future...
Yogesh Dahiya, G. Kaur, Neeraja Sahasrabudhe· 0 citations
We investigate the pathwise asymptotic dynamics of a nonlocal lattice equation with fixed delay on $\ell^2(\mathbb Z)$ driven by an infinite-dimensional stationary weakly geometric rough path. Using a rough-flow Doss-Sussmann transformation and the method of steps, we establish global well-posedness in the controlled-p...