Adaptive dataset reduction for neural networks
AssigneeNXP USA, Inc.
InventorsCurtis Hillier, Tristan Hayden McDaniel
Life of this patent
Estimating the term…
Abstract
Systems and methods for providing adaptive dataset reduction for neural networks. In an illustrative, non-limiting embodiment, a delta-sigma modulator may include a Systems-on-Chip (SoC) having a processor and a memory coupled to the processor, the memory having program instructions stored thereon that, upon execution, cause the processor to: reduce a dataset received from a sensor; and as the dataset is received, at least one of: in response to an anomaly in the dataset being present in the reduced dataset, indicate a pass; or in response to the anomaly in the dataset being absent from the reduced dataset, indicate an error.
Claims20
A Systems-on-Chip (SoC), comprising:
- a processor;
- and
- a memory coupled to the processor, the memory having program instructions stored thereon that, upon execution, cause the processor to:
- reduce a dataset received from a sensor;
- and
- as the dataset is received, at least one of:
- in response to an anomaly in the dataset being present in the reduced dataset, indicate a pass;
- or
- in response to the anomaly in the dataset being absent from the reduced dataset, indicate an error.
A hardware memory device having program instructions stored thereon that, upon execution by a processor, cause the processor to:
- identify a difference between an anomaly in a dataset being received from a sensor and an anomaly in a reduced dataset;
- and
- at least one of:
- in response to the difference being smaller than a threshold, indicate that the reduced dataset is adequate;
- or
- in response to the difference being greater than the threshold, indicate that the reduced dataset is inadequate.
A method, comprising:
- reducing a neural network training dataset captured by a sensor in real-time;
- detecting an anomaly in the dataset;
- and
- determining if the reduced dataset is representative of the dataset, at least in part, by comparing the anomaly against the reduced dataset.
Dependencies are inferred from the claim text.