US 12,596,961 B1Granted

Local device embeddings for automation

AssigneeAmazon Technologies, Inc.

InventorsSven Eberhardt, Amir Salimi, Jin Long Lee, Maisie Wang, Akanksha Gupta, Kaustubh Anilkumar Vibhute, Biwei Tao, Caglar Iskender

Life of this patent

Estimating the term…

patentgrep

Loading patent…

Abstract

Devices and techniques are generally described for local device embeddings for automation. In various examples, first data representing first state change data for network-connected computing devices configured in communication with a first network may be determined. The first data may be input into a first machine learning model. In some examples, the first machine learning model may generate first embedding data representing a combination of the first data and second data. In some examples, the second data may represent historical state change data for the network-connected devices. In some examples, the first embedding data may be stored in memory. A first action may be performed by a first network-connected device based at least in part on the first embedding data.

Claims20

Dependent claims

A method comprising:

determining, for a first time period, first data representing a change in a state of at least one network-connected computing device configured in communication with a network;
determining, for the first time period, second data representing a device crossing a perimeter defined by a geofence;
generating a first vector representing the first data and the second data using an encoder, wherein generation of the first vector is triggered by the device crossing the perimeter defined by the geofence;
sending the first vector as an input into a recurrent neural network (RNN);
generating, by the RNN, first embedding data representing a combination of the first vector at a current time step and hidden state data comprising a representation previous inputs to the RNN, wherein the hidden state data was generated during a previous time step prior to the current time step, and wherein the hidden state data represents a previous state of the at least one network-connected computing device;
determining, after generating the first embedding data, a triggering event associated with performance of a first action by a first network-connected computing device;
generating feature data describing the triggering event, wherein the feature data is used as input by a task-specific machine learning model associated with the first action, wherein the task-specific machine learning model is different from the RNN;
inputting the feature data and the first embedding data into the task-specific machine learning model associated with the first action;
and
generating output data by the task-specific machine learning model effective to cause the first network-connected computing device to perform the first action.

7 more elements

A system comprising:

at least one processor;
and
non-transitory computer-readable memory storing instructions that, when executed by the at least one processor, are effective to:
determine first data representing first states of computing devices configured in communication with a first network;
input the first data into a first trained machine learning model;
generate, by the first trained machine learning model, first embedding data representing a combination of the first data and second data, the second data representing historical state data for the computing devices;
store the first embedding data in persistent memory;
determine, using at least one of the computing devices, that a first event has occurred;
receive, from the persistent memory, the first embedding data based at least in part on the first event having occurred;
and
perform a first action by a first computing device based at least in part on the first embedding data and the first event.

8 more elements

Dependencies are inferred from the claim text.