The Whale That Outswam Evolution: Swarm Intelligence Maximises Memory in Connectome Reservoirs

Researchers applied four bio-inspired optimization algorithms to connectome-based neural networks across six animal species, demonstrating that gradient-free optimization can enhance biological neural structures by up to 17x on memory capacity tasks. The findings show that biological weight values, refined through evolution, serve as critical initial conditions that topology alone cannot replicate, establishing a principled approach for improving connectome-based reservoir computing systems.

This research bridges neuroscience and machine learning by testing whether evolutionary-shaped neural architectures can be systematically improved through optimization. The study evaluated four gradient-free algorithms—Particle Swarm Optimization, Differential Evolution, Grey Wolf Optimizer, and Whale Optimization Algorithm—on connectomes ranging from C. elegans' 279 neurons to human structural MRI connectivity. The Whale Optimization Algorithm consistently delivered the strongest results, achieving 214% average improvements across all species and benchmarks.

The findings challenge a fundamental assumption in neuroscience: that evolution has already optimized neural structures for their computational roles. Instead, the research reveals that biological connectomes encode valuable but suboptimal weight configurations. The critical insight emerges from comparison experiments where random weight initialization on the same topology underperforms biological baselines, proving that evolutionary weight refinement—not merely topology—carries essential computational information. This distinguishes genuine biological optimization from lucky architectural patterns.

For artificial intelligence development, these results suggest that biologically-inspired optimization strategies applied to learned neural architectures could yield significant performance gains. The consistency across six orders of magnitude in neural complexity indicates the approach generalizes robustly. The dramatic improvements in memory capacity and prediction accuracy on canonical benchmarks demonstrate practical relevance beyond theoretical interest.

Future research should investigate whether these optimization techniques transfer to artificial neural networks trained on real-world tasks, and whether the optimized connectome weights reveal design principles that evolution discovered independently. Understanding how evolution achieves near-optimal but improvable solutions could inform both neuroscience models and next-generation AI architectures.

• →Whale Optimization Algorithm improved connectome memory capacity by up to 17x, with 214% average gains across all species and tasks
• →Biological weight values from evolution serve as essential inductive biases that random initialization cannot recover despite identical topology
• →Bio-inspired optimization outperformed unoptimized connectomes across all six species (C. elegans through human) on four benchmark tasks
• →Results demonstrate evolution creates suboptimal but improvable neural structures, not mathematically optimal configurations
• →Findings suggest biologically-initialized optimization could enhance artificial neural networks trained on real-world applications