ANGLE: Angular Neural Generative Learning via Engression

ANGLE is a deep generative framework for non-parametric distributional regression on circular data, introduced to address limitations of traditional regression with angular responses. It learns full conditional distributions through a generalized circular energy score loss, with established theoretical properties including strict propriety and rotational equivariance. The framework handles extrapolation, dimension reduction, and distribution equality testing, demonstrating superior performance in object pose estimation and wind direction prediction tasks.