Akhil Perincherry


I am an ML research engineer at Ford Motor Company where I work on computer vision and machine learning for perception features in the context of automated driving. Most of my work is on camera images and LiDAR point clouds.

In my free time I enjoy playing/watching soccer, kickboxing, hiking (waterfall hikes are the best!) and practically any outdoor sport.

Email | LinkedIn | Github
Google Scholar | Twitter

Contrastive Training for Improved Out-of-Distribution Detection - Paper Summary


Existing techniques



CLP score


They showed that representations obtained through contrastive training improve OOD detection performance beyond what is possible with purely supervised training. The representations are shaped by joint training, in which the contrastive loss pushes the representations apart, even within each class, while the supervised loss acts to cluster the representations by class.