 |
deepTACOMA is an enhancement of the TACOMA algorithm for the scoring of TMA images. It was inspired by recent advances in deep representation learning algorithm and semi-supervised learning. It learns deep latent cluster structures in the data, and turn such information into a readily deployable representation--regularizing features--to be used with existing features.
The learning of the deep latent structures is empowered by a versatile tool we recently developed--rpForests. Our experiments show that rpForests is effective in extracting the cluster structure in the data. It is expected that our approach would work in settings where the labels are noisy and there are cluster structures in the data, or some weakly supervised settings. |