Author here. I pulled the accepted-paper lists from the four big ML/CV venues for 2024 and 2025, embedded title + abstract, ran UMAP to 2D, then HDBSCAN for clusters. 26,741 papers, 509 clusters, labeled by their most distinctive title words. Self-contained HTML with Plotly, no backend.
