-
This paper introduces a CNN framework where each CNN handles a particular task associated with an attribute. The latent representations generated by each of the CNN is combined into one shared latent feature matrix,
, that allows sharing of attribute knowledge among similar groups for better prediction.
-
The main contribution of the authors is the using the implementation of the latent task matrix to generated a shared latent matrix with all the different attribute features combined and capable of sharing knowledge among similar groups.
- The paper is heavily inspired from Zhou et.al that discusses the latent task matrix generation.
- One of the key assumptions in the paper is that sharing of knowledge amongst similar groups should be encouraged and discouraged among dissimilar groups.