Insight: Unsupervised Clustering of Quantum Complexity
#2
by QSBench - opened
Discovering Structural "DNA"
We’ve performed an unsupervised K-Means clustering on this dataset using structural features like gate_entropy, adj_density, and depth.
Key Findings:
- Natural Grouping: Even without labels, quantum circuits form distinct "complexity tiers" when projected via PCA.
- The "Scale" vs "Density": Our PCA analysis shows that the horizontal axis often represents the total gate count, while the vertical axis captures the density of interactions.
- Outlier Detection: We identified unique circuit topologies that sit far away from standard benchmark "clouds."
Discussion:
Has anyone tried applying more advanced clustering (like HDBSCAN) or graph neural networks (GNNs) to this metadata to predict fidelity?
We'd love to see your visualizations or notebooks!