Conflict-Driven Hybridization: Learning When to Switch Between Agentic and Classical Operators

TL;DR: What if we could make an agent that learns when to use AVO and when to fall back on classical mutation/crossover, based on how well each approach is doing? By detecting and analyzing situations where AVO and traditional methods disagree or perform differently, we could train a meta-controller to dynamically select or blend operators, aiming for best-of-both-worlds performance.

Research Question: Can a conflict-aware meta-controller, trained to detect divergences between agentic and classical variation operators, improve search efficiency and solution quality in evolutionary optimization?

Hypothesis: A meta-controller that adaptively switches between agentic variation and classical operators in response to observed conflicts or stagnation can outperform either approach alone—especially in non-stationary or multi-modal search landscapes.

Experiment Plan: Implement an evolutionary search framework where both AVO and standard operators are available. Instrument the system to detect “conflict points”—discrepancies in solution quality, diversity, or improvement rates between the two operator classes (see Borzenkov et al., 2025; Hoorfar, 2025). Train a lightweight RL-based or rule-based meta-controller to choose the operator (or a stochastic blend) at each generation based on recent performance metrics and detected conflicts. Benchmark on both kernel optimization (as in Chen et al., 2026) and multi-objective logistics/network design tasks (Borzenkov et al., 2025; Akopov & Beklaryan, 2025). Analyze Pareto fronts, convergence rates, and the conditions under which hybridization is most advantageous.

References:

• Chen, T., Ye, Z., Xu, B., Ye, Z., Liu, T., Hassani, A., Chen, T., Kerr, A., Wu, H., Xu, Y., Chen, Y., Chen, H., Kane, A., Krashinsky, R., Liu, M., Grover, V., Ceze, L., Bringmann, R. A., Tran, J., Liu, W., Xie, F., Lightstone, M. F., & Shi, H. (2026). AVO: Agentic Variation Operators for Autonomous Evolutionary Search.
• Borzenkov, A. M., Pronin, C., Podberezkin, A. A., Ostroukh, A., & Shmonin, A. M. (2025). Research methods for multicriteria optimization of logistics based on hybrid evolutionary-predictive models. Transportation and Information Technologies in Russia.
• Hoorfar, A. (2025). An Efficient Hybrid of Evolutionary Programing and Particle Swarm Optimization for Mixed-Parameter Electromagnetic Design Problems. 2025 IEEE CNC-USNC-URSI North American Radio Science Meeting (Joint with AP-S Symposium).
• Akopov, A., & Beklaryan, L. (2025). Evolutionary Synthesis of High-Capacity Reconfigurable Multilayer Road Networks Using a Multiagent Hybrid Clustering-Assisted Genetic Algorithm. IEEE Access.