US 11,861,490 B1
Decoupled machine learning training
Saurabh Gupta, Redmond, WA (US); Bharathan Balaji, Seattle, WA (US); Leo Parker Dirac, Seattle, WA (US); Sahika Genc, Mercer Island, WA (US); Vineet Khare, Redmond, WA (US); Ragav Venkatesan, Seattle, WA (US); and Gurumurthy Swaminathan, Redmond, WA (US)
Assigned to AMAZON TECHNOLOGIES, INC., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Nov. 21, 2018, as Appl. No. 16/198,726.
Int. Cl. G06N 3/08 (2023.01); G06N 3/04 (2023.01); G06F 18/214 (2023.01); G06F 18/21 (2023.01)
CPC G06N 3/08 (2013.01) [G06F 18/214 (2023.01); G06F 18/2178 (2023.01); G06N 3/04 (2013.01)] 21 Claims
OG exemplary drawing
 
1. A system for implementation of a reinforced machine learning environment, the system comprising:
a plurality of computing devices corresponding to a set of independently hosted customer environments, wherein individual independently hosted customer environments include a convolutional neural network agent process for generating and transmitting training data, wherein the training data is embodied as state information of the independently hosted customer environment, action information of the independently hosted customer environment, reward information of the independently hosted customer environment, and observation information of the independently hosted customer environment; and
one or more computing devices associated with a reinforced learning-based machine learning cluster, wherein the reinforced learning-based machine learning cluster is configured to:
receive a request from a customer computing device to provide a trained machine learning model that is trained in accordance with a reinforced learning model on training data obtained from a first independently hosted customer environment;
obtain the training data from the first independently hosted customer environment via an interface from one or more convolutional neural network agents associated with the first independently hosted customer environment;
train a machine learning model on the training data in accordance with the reinforced learning model to form the trained machine learning model; and
provide, to the customer computing device, the trained machine learning model or access to the trained machine learning model in response to the request.