Abstract
This chapter introduces the Continuous Restricted Boltzmann Machine, a probabilistic neural algorithm which is both useful in modelling continuous data and amenable to VLSI implementation. The capabilities of the model are explored with both artificial and real data. The computing units (neurons) and the unsupervised training rule have been implemented in VLSI. These results demonstrate the feasibility of a full VLSI model that uses continuous probabilistic behaviour to model the noise associated with all real signals, and therefore acts as a robust classifier or novelty detector.