Compositional Generalization Requires Linear, Orthogonal Representations in Vision Embedding Models
Paper โข 2602.24264 โข Published โข 14
Logbook, paper, scaled embeddings, poster, and rerunnable artifacts for the paper 805 reproduction.
Note Original ICML 2026 paper
Explore research logbook and sync findings with your agent
Note Canonical Trackio reproduction logbook
Note Scaled embeddings, metrics, complete bundle, and poster exports
Note Trackio artifact manifest and immutable bundle blob