Search for a command to run...
Metadata Title: "NEVO: A Neuromorphic EVolutionary Optimiser with Spike-Driven Cortico-Basal-Thalamic Coordination" Status: Accepted for Presentation at the IEEE WCCI/CEC 2026 (21-26 June 2026) Overview This repository contains the complete research implementation, experimental data, and analysis scripts for NEVO, a neuromorphic evolutionary optimiser that coordinates search operators through spike-driven neural circuits. Operator selection is implemented as action gating within a Cortico-Basal-Thalamic Loop (CBTL) whilst maintaining standard black-box optimisation interfaces. Evaluation on the BBOB noiseless benchmark suite (2,160 runs across 24 functions, 6 dimensions, 15 instances each) identifies reward design as the primary scalability constraint in neuromorphic operator coordination. Key Contributions 1. First demonstration of spike-driven operator coordination in evolutionary optimisation 2. Systematic identification of reward design as the primary scalability barrier 3. Open-source implementation compatible with neuromorphic hardware (Nengo, extensible to Loihi 2, SpiNNaker) 4. Comprehensive experimental analysis across 2,160 BBOB runs (24 functions × 6 dimensions × 15 instances) Reproducibility • All code, configurations, and raw experimental data included • Complete step-by-step instructions for running benchmarks Repository Structure nevo-main.zip Complete Python implementation of the NEVO framework. Contains core optimiser logic, thirteen metaheuristic operators, Nengo-based CBTL neural circuit, state feature extraction, and benchmark scripts. The basic_example.py script generates preliminary validation figures while benchmark_experiment.py executing the full BBOB suite evaluation. preliminary-figures.zip Visualisations generated by nevo/examples/basic_example.py for the first experiment set. Shows single-run dynamics on representative BBOB functions (f01, f06, f10, f15, f20) in 10D with 5-second simulation time. ppdata.zip Post-processed analysis results derived from exdata.zip. Includes BBOB 2009 baseline algorithm data (BIPOP-CMA-ES, BFGS, DE-PSO, RANDOMSEARCH), target attainment summaries for ECDF comparisons, and many more resources. exdata.zip Raw experimental data from nevo/examples/benchmark_experiment.py executing the complete BBOB noiseless suite. Contains 2,160 individual runs (24 functions × 6 dimensions × 15 instances) with CocoEx format fitness trajectories and experiment metadata. cocoex-complementary.zip COCO benchmark complementary data generated alongside exdata.zip by benchmark_experiment.py. Contains aggregated performance metrics, operator weight evolution data, success rate statistics, and publication-ready figures (Figures 3-6 in manuscript). Includes CSV files for quantitative analysis and PNG/PDF vector graphics for manuscript figures.