A sparse contrastive reduction is even further proposed to train our product with sparse supervision. Our new formulation generates a 3D heatmap indicating the plausibility of all location/scale mixtures in a single network ahead go, which happens to be more than ten periods faster in comparison to the past sliding-window system. Additionally, it s