neuroptimiser 1.0.3

A modular and asynchronous framework for neuromorphic optimisation using spike-driven metaheuristics and heuristic-controlled spiking dynamics.

NeurOptimiser

NeurOptimiser is a neuromorphic optimisation framework in which metaheuristic search emerges from asynchronous spiking dynamics. It defines optimisation as a decentralised process executed by interconnected Neuromorphic Heuristic Units (NHUs), each embedding a spiking neuron model and a spike-triggered heuristic rule.

This framework enables fully event-driven, low-power optimisation by integrating spiking computation with local heuristic adaptation. It supports multiple neuron models, perturbation operators, and network topologies.

---

✨ Key Features

• Modular and extensible architecture using Intel’s Lava.
• Supports linear and Izhikevich neuron dynamics.
• Implements random, fixed, directional, and Differential Evolution operators as spike-triggered perturbations.
• Includes asynchronous neighbourhood management, tensor contraction layers, and greedy selectors.
• Compatible with BBOB (COCO) suite.
• Designed for scalability, reusability, and future deployment on Loihi-class neuromorphic hardware.

---

📖 Documentation

For detailed documentation, examples, and API reference, please visit the Neuroptimiser Documentation.

📦 Installation

pip install -e .

Ensure you have Python ≥ 3.10 and the Lava-NC environment configured.

You can also clone the repository and install it. Check the Makefile for additional options.

🚀 Example Usage

from neuroptimiser import NeuroOptimiser
import numpy as np

problem_function    = lambda x: np.linalg.norm(x)
problem_bounds      = np.array([[-5.0, 5.0], [-5.0, 5.0]])

optimiser = NeurOptimiser()

optimiser.solve(
    obj_func=problem_function,
    search_space=problem_bounds,
    debug_mode=True,
    num_iterations=1000,
)

For more examples, please, visit Neuroptimiser Usage

📊 Benchmarking

Neuroptimiser has been validated over the BBOB suite, showing:

• Competitive convergence versus Random Search
• Consistent results across function types and dimensions
• Linear runtime scaling with number of units and problem size

🔬 Citation

@misc{neuroptimiser2025,
  author={Cruz-Duarte, Jorge M. and Talbi, El-Ghazali},
  title        = {Neuroptimiser: A neuromorphic optimisation framework},
  year         = {2025},
  url          = {https://github.com/neuroptimiser/neuroptimiser},
  note         = {Version 1.0.X, accessed on 20XX-XX-XX}
}

🔗 Resources

• 📘 Documentation
• 📜 Paper
• 🧠 Intel Lava-NC
• 🧪 COCO Platform

🛠️ License

MIT License — see LICENSE

🧑‍💻 Authors

• Jorge M. Cruz-Duarte — University of Lille
• El-Ghazali Talbi — University of Lille