Atomic Vector Symbolic Architecture: Physics-Inspired Hyperdimensional Computing for Explainable AI

This study introduces the Atomic Vector Symbolic Architecture (Atomic VSA), a deterministic framework grounded in Hyperdimensional Computing (HDC) for clinical triage and explainable AI.

Key Results:
• 92.5% F1 Score on 25-category ICD-11 clinical triage (Winner-Take-All, no tuning)
• 91.9% label recall for multi-label comorbidity detection
• 11.97ms median inference latency on commodity CPU (15W)
• Algebraically-traceable decisions enabling full audit trails

The system employs 10,048-dimensional bipolar vectors with holographic reduced representations, enabling semantic composition through binding and bundling operations. We document an 8% accuracy ceiling due to semantic clones with 100% symptom overlap—a fundamental symptom-encoding limit requiring lab values for disambiguation.

Atomic VSA is positioned as a complementary paradigm to neural networks for scenarios prioritizing deterministic inference, on-premise deployment, and regulatory compliance.