Skip to Main Content
Novel reversible neural network (RevNN) architecture is introduced, and a RevNN paradigm using supervised learning is presented. The application of RevNN to multiple-output feedforward plant control is shown. (k,k) reversible circuits are circuits that have the same number of inputs (k) and outputs (k) and are one-to-one mappings between vectors of inputs and outputs, thus the vector of input values can always be uniquely reconstructed from the vector of output values. Since the reduction of power consumption is a major requirement for the circuit design of future technologies such as in quantum computing, the main features of several future technologies will include reversibility, and thus the new RevNN circuits can play an important role in the design of circuits that consume minimal power for applications such as low-power control of autonomous robots.