This repository explores Formal Concept Analysis (FCA) and its integration with optimization models (QUBO) to tackle problems in classification and link prediction.
The project is organized into two main branches:
- JSM Classification → Optimized concept generation and classification using Iceberg FCA and JSM method.
- Link Prediction → Flexible biclique dataset selection and optimization using FCA lattices.
The main branch contains only this README as an overview.
The project investigates how FCA can be combined with classical and quantum optimization techniques to:
- Generate and prune concept lattices efficiently.
- Apply formal concepts for classification through JSM hypothesis generation.
- Select bicliques in datasets for improved link prediction.
- Design QUBO formulations to balance predictive quality, simplicity, and coverage.
- Benchmark classical approaches against optimization-based ones.
This branch applies FCA to build and prune concept lattices, then uses them for classification tasks.
- Data is encoded into FCA-ready structures.
- Concepts are generated and filtered with support thresholds (Iceberg approach).
- Positive hypotheses are constructed following the JSM method.
- Optimization is framed as a QUBO problem balancing predictive gain and rule complexity.
- Compatible with both classical solvers and quantum-inspired solvers.
This branch focuses on link prediction using biclique selection within concept lattices.
- Concepts are generated via brute-force and Next Closure approaches.
- Hasse diagrams are used to visualize lattice structures.
- Biclique selection is formulated as an optimization problem.
- A QUBO model balances object coverage with feature importance, showcasing flexibility in lattice-driven optimization.
- Clone the repository:
git clone https://github.com/iCog-Labs-Dev/Quantum-optimization-for-fca cd Quantum-optimization-for-fca
git checkout jsm-classification
git checkout link-prediction