Fig. 2From: Unsupervised deep learning reveals prognostically relevant subtypes of glioblastomaDeep belief network (DBN) with two phases: pre-training and fine-tuning. Each RBM in the pre-training phase iteratively learns lower dimensional representations of the input (gene expression microarray) one RBM at a time. These lower dimensional representations are then used as input to the next RBM. The pre-training phase learns a set of weights (WÂ =Â w 1 , w 2 , and w 3 ) that is used to initialize the fine-tuning phase. The fine-tuning phase optimizes the weights by minimizing the cross-entropy errorBack to article page