- Adaptive control system and methods for training artificial neural networks (ANNs)
- Enables the use of parameters and intermediary variables with adjustable ranges
- Minimizes errors in the target output by applying probability distributions to parameters and associated ranges during the adaptation process
The University of Central Florida invention describes an adaptive system with parameters and intermediary variables with adjustable ranges. Artificial neural networks (ANN) are a class of adaptive systems that transform inputs such as text, audios, images, videos, to outputs of desired formats to achieve various tasks. The ranges are determined based on the statistics of the parameters/variables during the adaptation process to reduce errors caused by clipping and/or low precision, leading to improved performance in range-limited adaptive systems. During the adaptation process, the range limitations impose a constraint on system parameters and intermediary parameters, which may compromise the performance of the system.