MFEA-CoD shifts evolution from consistency to collaborative novelty discovery

Evolutionary multitasking (EMT) has long focused on objective-driven optimization, exploiting inter-task consistency to speed up convergence toward predefined optima. In a new arXiv paper, Jiao Liu, Yanchi Li, Hua Yu, Abhishek Gupta, and Yew-Soon Ong break from this paradigm by introducing MFEA-CoD—a multifactorial evolutionary algorithm built for collaborative discovery in novelty search. Instead of merely transferring consistent search information, MFEA-CoD coordinates multiple novelty search tasks to jointly explore behaviorally novel regions of the solution space. Two key innovations drive this shift: a multitask repulsion operator that pushes tasks into distinct subregions to avoid redundant behavioral discoveries, and an adaptive inter-task transfer mechanism that dynamically adjusts transfer probability based on the online contribution of shared information. The algorithm is further extended to multitask novelty-augmented optimization, where behavioral novelty and objective information are combined to prevent premature convergence in deceptive landscapes.

The team evaluated MFEA-CoD across four challenging problem types: synthetic basin-type problems, deceptive maze navigation, MuJoCo policy optimization, and generative novelty search. Results show that the algorithm significantly improves the efficiency of discovering diverse novel solutions compared to standard EMT baselines. The repulsion operator ensures broader exploration without sacrificing quality, while the adaptive transfer mechanism prevents over-sharing of misleading information. In deceptive objective landscapes—where traditional objective-driven methods often get stuck—MFEA-CoD maintains diversity and avoids local optima. This work opens a new direction for EMT: moving from consistency-based acceleration to collaborative, novelty-focused discovery. The paper is available on arXiv (2607.00761) and could have implications for robotics, game AI, and any domain requiring creative exploration.