Xavier Bou

Early transformer layers act as in-context PCA engines, compressing high-dimensional visual demonstrations into a universal, low-dimensional subspace where classification becomes trivially separable.

Season-invariant feature matching can drastically cut down geometric errors in satellite imagery, making GCP-free refinement more reliable than ever.

Forget wearables: broadcast footage, processed with game state reconstruction and novel kinematics, offers a surprisingly viable, low-cost path to tracking athlete fatigue.