MOSAIC: Modular Open Symbolic Architecture for Interpretable Classification
Jesús Manuel Soledad Terrazas · 2025
We present MOSAIC (Modular Open Symbolic Architecture for Interpretable Classification),a four-component framework that maintains full interpretability while achieving high classification performance. MOSAIC integrates feature extraction, explicit knowledge representation,weighted reasoning, and incremental learning into a unified architecture. We demonstrate theframework through two distinct mathematical problems: prime number classification (achieving100% accuracy on datasets up to 10 million integers) and Collatz sequence halting prediction(effective learning with 0.5% labeled data). Both implementations maintain complete decisiontraceability. The modular design enables independent component development and provides afoundation for future scalability research. This work establishes a baseline architectural patternfor building interpretable AI systems with clear pathways for systematic enhancement