# US 12,010,178 B2 · Intelligent edge computing platform with machine learning capability

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,010,178 B2
- Document ID: US12010178B2
- Status: Granted
- Kind: B2
- Application number: 18/125,954
- Filed: 2023-03-24
- Published: 2023-09-21
- Granted: 2024-06-11
- Earliest claimed: 2019-04-09
- Assignees: Tyco Fire & Security GmbH
- Applicants: Tyco Fire & Security GmbH
- Inventors: Abhishek Sharma; Sastry K M Malladi
- Web page: https://patentgrep.com/patent/US12010178B2

## CPC

- H04L 67/12: specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks
- G06F 18/2148: {characterised by the process organisation or structure, e.g. boosting cascade}
- G06F 18/251: {of input or preprocessed data}
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06N 5/04: Inference or reasoning models
- G06N 20/00: Machine learning
- G16Y 40/20: Analytics; Diagnosis
- H04L 67/10: in which an application is distributed across nodes in the network (software deployment G06F8/60; multiprogramming arrangements G06F9/46)
- H04L 67/34: {involving the movement of software or configuration parameters (network booting or remote initial program loading [RIPL] G06F9/4416)}
- H04L 67/562: Brokering proxy services
- H04L 67/566: Grouping or aggregating service requests, e.g. for unified processing
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/045: Combinations of networks
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N 20/10: using kernel methods, e.g. support vector machines [SVM]
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G16Y: INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]

## Abstract

An edge computing platform with machine learning capability is provided between a local network with a plurality of sensors and a remote network. A machine learning model is created and trained in the remote network using aggregated sensor data and deployed to the edge platform. Before being deployed, the model is edge-converted (“edge-ified”) to run optimally with the constrained resources of the edge device and with the same or better level of accuracy. The “edge-ified” model is adapted to operate on continuous streams of sensor data in real-time and produce inferences. The inferences can be used to determine actions to take in the local network without communication to the remote network. A closed-loop arrangement between the edge platform and remote network provides for periodically evaluating and iteratively updating the edge-based model.

## Claims

### Claim 1

A method comprising:
providing an edge platform connected to a local network between a plurality of sensors and a remote network;
using an edge converter creating an edge-converted machine learning model, wherein the edge-converted machine learning model comprises analytics expressions of an analytics expression language that are operable to process streams of data in real-time;
providing the edge-converted machine learning model on the edge platform;
receiving a first sensor data stream from a first sensor of the plurality of sensors at the edge platform;
at the edge platform without first transmitting the first sensor data stream to the remote network for processing:
executing the edge-converted machine learning model at least partially in a complex event processing engine to operate on the first sensor data stream in real-time and produce a stream of first inferences about a first network device of a plurality of network devices; and
determining based on the stream of first inferences whether to take an action regarding the first network device without first communicating the inferences to the remote network.

### Claim 2 (depends on 1)

The method of claim 1, wherein creating the edge-converted machine learning model comprises:
converting the non-edge-based machine learning model to an intermediate form adapted for deployment; and
translating the intermediate form of the non-edge-based machine learning model into a script of the analytics expressions of the analytics expression language.

### Claim 3 (depends on 1)

The method of claim 1, wherein:
the non-edge-based machine learning model is operable to recognize particular semantics in the aggregated static sensor data; and
the edge-converted machine learning model is operable to recognize the same particular semantics in the streaming sensor data in real-time.

### Claim 4 (depends on 1)

The method of claim 1, wherein:
the non-edge-based machine learning model comprises a data pre-processing component; and
creating the edge-converted machine learning model comprises removing the data pre-processing component and replacing it with equivalent expressions of an analytics expression language that are executable in the complex event processing engine separately from the edge-converted machine learning model.

### Claim 5 (depends on 4)

The method of claim 4, wherein:
the non-edge-based machine learning model comprises a data post-processing component; and
creating the edge-converted machine learning model comprises removing the data post-processing component and replacing it with equivalent expressions of the analytics expression language that are executable in the complex event processing engine separately from the edge-converted machine learning model.

### Claim 6 (depends on 1)

The method of claim 1, wherein creating the edge-converted machine learning model comprises:
parsing the non-edge-based machine learning model and identifying computations it contains for the aggregated static data;
building in a memory a computation flow graph representation including the identified computations;
optimizing the identified computations and the computation flow represented in the computation flow graph for real-time operation on the streaming sensor data; and
incorporating the optimized computations and computation flow in the edge-converted machine learning model.

### Claim 7 (depends on 1)

The method of claim 1, wherein creating the edge-converted machine learning model comprises:
verifying the edge-converted machine learning model produces accurate results compared to the non-edge-based machine learning module for a given set of data values.

### Claim 8 (depends on 1)

The method of claim 1, wherein the verifying comprises:
parsing model verification tags from the non-edge-based machine learning module;
extracting from the non-edge-based machine learning module a first set of input data
values and a first set of expected output data values corresponding to the verification tags;
converting the first sets of input data values and expected output data values to a second set of input data values and a second set of expected output data values respectively in a form compatible with the edge-converted machine learning model;
executing the edge-converted machine learning model on the second set of input data values and producing a set of actual output data values; and
comparing the set of actual output data values and the second set of expected output data values.

### Claim 9 (depends on 1)

The method of claim 1, wherein:
the remote network comprises a cloud manager;
the edge platform comprises an edge manager; and
providing the edge-converted machine learning model on the edge platform comprises the edge manager receiving the edge-converted machine learning model directly from the cloud manager.

### Claim 10 (depends on 1)

The method of claim 1, comprising:
providing an app store intermediate the remote network and the edge platform;
wherein providing the edge-converted machine learning model on the edge platform comprises transferring the edge-converted machine learning model from the remote network to the app store and retrieving the edge-converted machine learning model from the app store to the edge platform.

### Claim 11 (depends on 1)

The method of claim 1, comprising:
in response to a trigger, transmitting one or more inferences from the stream of first inferences and one or more data values from the first sensor data stream to the remote network;
applying the non-edge based machine learning model to the one or more data values and producing one or more second inferences;
comparing the one or more second inferences with the one or more inferences from the first stream of inferences;
based on the comparison, generating a modified edge-converted machine learning model; and
updating the edge-converted machine learning model on the edge platform with the modified edge-converted machine learning model.

### Claim 12 (depends on 11)

The method of claim 11, comprising:
executing expressions of an analytics expression language in the complex event processing engine on the stream of first inferences;
generating over time a characteristic of the first inferences;
comparing the characteristic to a value corresponding to a predetermined level of unacceptable accuracy of the edge-converted machine learning model; and
creating the trigger based on the comparing.

### Claim 13 (depends on 11)

The method of claim 11, wherein updating the edge-converted machine learning model with the modified machine learning model comprises:
generating a continuous model stream representing the edge-converted machine learning model;
streaming the model stream along with the first sensor data stream on the edge platform;
receiving at the edge computing platform the modified edge-converted machine learning model;
inserting the modified edge-converted machine learning model into the model stream;
detecting a change in the model stream as the modified edge-converted machine learning model propagates in the model stream;
in response to detecting the change in the model stream instantiating an instance of the modified edge-converted machine learning model; and
using the instantiated instance of the modified edge-converted machine learning model to operate on the first sensor data stream and produce the stream of first inferences in place of the edge-converted machine learning model.

### Claim 14 (depends on 1)

The method of claim 1, comprising:
using the edge converter creating a plurality of edge-converted machine learning models operable in real-time on the streaming sensor data from a plurality of non-edge-based machine learning models trained and operable in non-real-time on aggregated static sensor data;
wherein each of the plurality of non-edge-based machine learning models includes a different type of machine learning model represented in a different high-level model language;
wherein creating the plurality of edge-converted machine learning models includes converting each of the plurality of non-edge-based machine learning models to a script of expressions of the same analytics expression language adapted for execution in the complex event processing engine of the edge platform on the streaming sensor data in real-time; and
providing the plurality of edge-converted machine learning models on the edge platform.

### Claim 15

A method comprising:
providing an edge platform connected to a local network between a plurality of sensors and a remote network;
using an edge converter to create an edge-converted machine learning model from a non-edge-based machine learning model that is trained and operable in non-real-time on aggregated static sensor data, wherein the edge-converted machine learning model comprises analytics expressions of an analytics expression language that are operable to process streams of data in real-time;
providing on the edge platform the edge-converted machine learning model that is optimized to operate with the computing resources of the edge platform in real-time on one or more streams of sensor data and produce one or more streams of inferences;
receiving a first sensor data stream from a first sensor of the plurality of sensors at the edge platform;
without first transmitting the first sensor data stream to the remote network for processing:
executing the edge-converted machine learning model using the computing resources of the edge platform to operate on the first sensor data stream in real-time and produce a stream of first inferences about a first network device of the plurality of network devices; and
determining based on the stream of first inferences whether to take an action regarding the first network device without first communicating the inferences to the remote network.

### Claim 16

A method comprising:
providing on an edge platform of a distributed network an edge-converted machine learning model adapted to operate on a continuous stream of data in real-time and to produce a continuous stream of inferences in real-time;
receiving at the edge platform a first sensor data stream from a first sensor;
at the edge platform without first transmitting the first sensor data stream to the remote network for processing:
operating on the first sensor data stream with the edge-converted machine learning model and producing a stream of first inferences about a first network device; and
determining based on the stream of first inferences whether to take an action affecting the first network device;
updating the edge-converted machine learning model on the edge platform without interrupting the operating on the first sensor data stream or the producing the stream of first inferences, wherein the updating comprises:
generating a continuous model stream representing the edge-converted machine learning model;
streaming the model stream along with the first sensor data stream on the edge platform;
providing on the edge platform a modified edge-converted machine learning model; inserting the modified edge-converted machine learning model into the model stream;
detecting a change in the model stream corresponding to the modified edge-converted machine learning model as the modified machine learning model propagates in the model stream;
in response to detecting the change in the model stream instantiating an instance of the modified edge-converted machine learning model on the edge platform; and
going forward using the instantiated instance of the modified edge-converted machine learning model to operate on the first sensor data stream and produce the stream of first inferences in place of the edge-converted machine learning model.

### Claim 17 (depends on 16)

The method of claim 16, wherein detecting the change in the model stream and instantiating an instance of the modified edge-converted machine learning model comprise:
using a stateful subsystem to detect the change in the model stream in real-time;
in response to the stateful subsystem detecting the change, triggering an actor that understands the semantics of the model stream, wherein the actor:
reads the modified edge-converted machine learning model from the model stream and instantiates the instance of the modified edge-converted machine learning model;
couples the instantiated instance of the modified edge-converted machine learning model to the same semantics of the first sensor data stream as the edge-converted machine learning model; and
inserts the instantiated instance of the modified edge-converted machine learning model back into the model stream.

### Claim 18 (depends on 16)

The method of claim 16, wherein the edge platform has an edge manager, the remote network has a cloud manager, and providing the modified edge-converted machine learning model on the edge platform comprises the edge manager receiving the modified edge-converted machine learning model directly from the cloud manager.

### Claim 19 (depends on 16)

The method of claim 16, comprising an app store intermediate the edge platform and the remote network, and wherein providing the modified edge-converted machine learning model on the edge platform comprises retrieving the modified edge-converted machine learning model from the app store.