Adaptive training of neural network models at model deployment destinations
AssigneeApple Inc.
InventorsYichuan Tang, Nitish Srivastava, Ruslan Salakhutdinov
Life of this patent
Estimating the term…
Abstract
At a centralized model trainer, one or more neural network based models are trained using an input data set. At least a first set of parameters of a model is transmitted to a model deployment destination. Using a second input data set, one or more adaptive parameters for the model are determined at the model deployment destination. Using the adaptive parameters, one or more inferences are generated at the model deployment destination.
Claims20
A system, comprising:
- a model trainer comprising a first set of one or more computing devices;
- and
- one or more model deployment destinations, including a first model deployment destination comprising a second set of one or more computing devices;
- wherein the model trainer comprises instructions that when executed on one or more processors cause the first set of one or more computing devices to:
- train, using a first input data set comprising data collected from a plurality of sensors, a first neural network based model;
- transmit at least a first set of parameters of the first neural network-based model to the first model deployment destination;
- and
- wherein the first model deployment destination comprises instructions that when executed on one or more processors cause the second set of one or more computing devices to:
- determine, using a second input data set and an unsupervised learning algorithm, one or more adaptive parameters for the first neural network based model, wherein at least a subset of the first set of parameters remains unmodified by unsupervised learning;
- and
- generate, using a version of the first neural network based model comprising the one or more adaptive parameters, at least one inference.
A method, comprising:
- performing, by one or more computing devices:
- training, at a first model trainer, using a first input data set, one or more neural network based models;
- transmitting at least a first set of parameters of a first neural network-based model of the one or more neural network based models to a first model deployment destination;
- determining, at the first model deployment destination, using a second input data set and an unsupervised learning algorithm, one or more adaptive parameters for at least one model of the one or more neural network based models, wherein at least a subset of the first set of parameters remains unmodified by the unsupervised learning;
- and
- generating, at the first model deployment destination, using at least one model of the one or more neural network based models and the one or more adaptive parameters, at least one inference.
A non-transitory computer-readable storage medium storing instructions that when executed on one or more processors cause the one or more processors to:
- determine, at a model deployment destination, that at least a first set of parameters of one or more neural network based models has been received from a trainer resource, wherein at least a portion of training of the one or more neural network based models is performed at the trainer resource using a first input data set;
- obtain, at the model deployment destination, using a second input data set and an unsupervised learning algorithm, one or more adaptive parameters for at least one model of the one or more neural network based models, wherein at least a subset of the first set of parameters remains unmodified by the unsupervised learning;
- and
- generate, at the model deployment destination, using at least one model of the one or more neural network based models and the one or more adaptive parameters, at least one inference.
Dependencies are inferred from the claim text.