KANISA (meaning “think”) is a unified computational framework designed to support advanced research and applied development in:

• Global and local optimisation
• Numerical analysis
• Dynamic systems and control
• Machine learning
• Data-driven decision analytics

KANISA aims to provide a rigorous, extensible, and transparent environment for algorithmic development, benchmarking, and reproducible scientific research.

If you use KANISA in your research, please follow the citation guidelines below:

➡️ View the Complete Citation File

KANISA is organised into two primary components:

The main KANISA library is implemented in Python and provides modules for:

• Global optimisation algorithms
• Local search and numerical optimisation routines
• Interval and uncertainty methods (planned)
• Dynamic system modelling and control
• Machine learning utilities
• Decision intelligence and multi-criteria frameworks

This Python library is under active development and will evolve into the primary, fully supported KANISA environment.

KANISA includes a dedicated matlab/ directory preserving the original MATLAB implementations of key optimisation algorithms developed during earlier academic research phases. These versions serve as reference and archival implementations for transparency and reproducibility.

The MATLAB legacy component is divided into two categories:

These solvers are based on deterministic global optimisation and interval-based convex relaxations.

Algorithms included:

• αBB (alpha Branch-and-Bound)
• PSO–αBB Hybrid Algorithm

Dependency Notice:
These implementations rely on INTLAB, a proprietary interval arithmetic toolbox developed by Prof. S.M. Rump.
INTLAB is not included in this repository, and cannot be redistributed.
To run the rigorous versions, users must obtain a valid INTLAB licence from the official source.

These solvers represent the evolutionary computing lineage of KANISA.

Algorithms included:

• PSO (Particle Swarm Optimisation)
• DE (Differential Evolution)

These algorithms do not depend on proprietary software and are provided for reference, benchmarking, and historical documentation.

The MATLAB implementations are included for:

• Preserving the scientific and historical lineage of KANISA
• Ensuring reproducibility of published research
• Allowing researchers to replicate earlier experimental environments
• Providing authoritative baseline implementations for comparison with new Python solvers

Future versions of KANISA will provide pure Python equivalents of all legacy algorithms.