Deep Belief Networks (DBNs)

DBNs consist of multiple layers of interconnected units, typically built by stacking Restricted Boltzmann Machines (RBMs). Each layer learns increasingly abstract features from the output of the previous layer. DBNs can be trained layer by layer using unsupervised learning and then fine-tuned using supervised learning for specific prediction tasks.

DBNs helped demonstrate how deep neural networks could learn useful hierarchical representations from large datasets. Their layer-wise training approach also addressed some difficulties associated with training deep networks. Although newer architectures are now more common, DBNs played an important role in the development of modern deep learning.