Neural network applications in resource constrained environments

#This patent document relates to systems, devices, and processes that use neural networks in resource-constrained environments.

#Artificial neural networks are processing devices that are somewhat modeled after the neural structure of a cerebral cortex. Neural networks perform a type of supervised learning, using known inputs and known outputs to develop a model to categorize, classify, or predict something about a future input. Neural networks are highly adaptable and thus have been employed in many different domains.

#According to some embodiments of the present disclosure, a system is provided. The system includes a sensor located in a resource-constrained environment configured to generate sensor data of the resource-constrained environment. The system further includes a first computing device not located in the resource-constrained environment configured to produce a neural network structure based on the sensor data. The system further includes a second computing device located in the resource-constrained environment configured to provide the sensor data as input to the neural network structure. The second computing device is further configured to determine a state of the resource-constrained environment based on the input of the sensor data to the neural network structure.

#FIG. 1 is a diagram of a resource-constrained environment in which a neural network may be employed.

#FIG. 2 is a diagram of the interior of an automobile in which a neural network may be employed according to some embodiments of the present disclosure.

#FIG. 3 is an schematic diagram from an overhead view of an automobile in which a neural network may be employed according to some embodiments of the present disclosure.

#While neural networks are versatile and have been employed in many domains, there are significant challenges to deploying them in certain environments. This is because neural networks traditionally require significant processing power to train (i.e., to initially setup) and even to use.

#As with other types of machine learning, neural networks leverage large volumes of input data in order to learn patterns about an environment. Also as with other forms of machine learning, this process of learning patterns from large volumes of data (“training”) can require both significant data storage resources and significant computer processing resources. While efficient algorithms have been developed to perform this training of the neural network, the training nonetheless requires many steps of data processing. As such, significant processing power is often used to speed up this training, such as with multi-core processors, multiple computers operating concurrently, and other forms of high-powered computing.

#The output of the training process for a neural network is a set of configuration parameters that define the neural network and that reflect a model of the environment which the input data describes. The neural network defined by these configuration parameters can then be used with new input data in order to classify, categorize, or predict some value for the new input data.