CEC 2025 Competition on Black-box
Consensus-based Distributed Optimization

1、Competition Outline

Consensus-based Distributed optimization (CDO) is a common problem definition for optimization problems in networked systems. Usually, a networked system contains multiple physical or virtual entities, which are termed nodes or agents. In CDO, there is a local objective function for each node, and the systematic objective function, i.e. global objective function, is the sum of all local objective functions. CDO aims to minimize the global objective function and make the nodes reach a consensus on the final solution.

The goal of the competition is to encourage participants to use zero-order optimization algorithms such as evolutionary computation to improve performance of black-box CDO. To this end, we design a set of benchmark functions for black-box consensus-based distributed optimization. This benchmark set considers different communication environments, conflict degrees, node homogeneity, and types of objective functions. The main rule of this competition is to find the best possible global solution under the specified number of evaluations, while taking into account the consensus of the system and the communication efficiency. This competition is promising to encourage more related research and extend the application of evolutionary computation to real-world distributed and multi-agent systems.

2、Important Dates

Submission deadline: May 8, 2025
Notification (final ranking): June 8, 2025

Participants can submit the related code and results via emails to cstaiutan@mail.scut.edu.cn.

Offical website: Competition-on-Black-box-Consensus-based-Distributed-Optimization

3、Competition description

The competition provides an algorithm development platform for DCO. This platform provides interfaces for evaluation functions, communication, and performance evaluation, allowing developers to focus only on algorithm design. First, we design 5 groups of 36 benchmark functions in total for black-box and non-convex DCO, and provide evaluation interfaces for these functions. Besides, this competition considers a real-world application, the multi-target localization problem in wireless sensor networks. Second, we provide peer-to-peer communication interfaces based on the communication topology of benchmark functions. These interfaces Confirm that each node can only communicate with immediate neighbors. Third, we provide the performance evaluation interface for algorithms, including solution quality, communication efficiency, and system consensus. Framework, benchmark, and data are available in:
https://github.com/iamrice/Proposal-for-Competition-on-Black-box-and-Non-convex-Distributed-Consensus-Optimization