Machine learning techniques for precise position determination
AssigneeIntel Corporation
InventorsJeongsik Choi, Venkatesan Nallampatti Ekambaram, Yang-Seok Choi
Life of this patent
Estimating the term…
Abstract
Systems, methods, computer program products, and apparatuses to determine, by a neural network based on training data related to wireless signals exchanged by a device and a plurality of wireless access points in an environment, a respective distance between the device and each wireless access point, receive location data related to a respective location of each wireless access point of the plurality of wireless access points, determine a geometric cost of the neural network based on a geometric cost function, the respective distances, and the received location data, and train a plurality of values of the neural network based on a backpropagation and the determined geometric cost.
Claims17
An apparatus, comprising:
- a processor circuit;
- and
- memory storing instructions which when executed by the processor circuit cause the processor circuit to:
- determine, by a neural network based on training data related to wireless signals exchanged by a device and a plurality of wireless access points in an environment, a respective distance between the device and each wireless access point of the plurality of wireless access points;
- receive location data related to a respective location of each wireless access point of the plurality of wireless access points;
- determine a geometric cost of the neural network based on a geometric cost function, the respective distances, and the received location data, the geometric cost function to determine the geometric cost based on a two-dimensional coordinate of each wireless access point specified in the received location data, the respective distances, and a reliability value associated with each respective distance, the reliability value to comprise a standard deviation value;
- and
- train a plurality of values of the neural network based on a backpropagation and the determined geometric cost.
A non-transitory computer-readable storage medium storing instructions which when executed by a processor circuit cause the processor circuit to:
- determine, by a neural network based on training data related to wireless signals exchanged by a device and a plurality of wireless access points in an environment, a respective distance between the device and each wireless access point of the plurality of wireless access points;
- receive location data related to a respective location of each wireless access point of the plurality of wireless access points;
- determine a geometric cost of the neural network based on a geometric cost function, the respective distances, and the received location data, the geometric cost function to determine the geometric cost based on a two-dimensional coordinate of each wireless access point specified in the received location data, the respective distances, and a reliability value associated with each respective distance, the reliability value to comprise a standard deviation value;
- and
- train a plurality of values of the neural network based on a backpropagation and the determined geometric cost.
A method, comprising:
- determining, by a neural network executing on a processor based on training data related to wireless signals exchanged by a device and a plurality of wireless access points in an environment, a respective distance between the device and each wireless access point of the plurality of wireless access points;
- receiving location data related to a respective location of each wireless access point of the plurality of wireless access points;
- determining a geometric cost of the neural network based on a geometric cost function, the respective distances, and the received location data, the geometric cost function to determine the geometric cost based on a two-dimensional coordinate of each wireless access point specified in the received location data, the respective distances, and a reliability value associated with each respective distance, the reliability value to comprise a standard deviation value;
- and
- training a plurality of values of the neural network based on a backpropagation and the determined geometric cost.
Dependencies are inferred from the claim text.