# US 2021/0279567 A1 · Method and system for adjusting a machine learning output

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

- Number: US 2021/0279567 A1
- Document ID: US20210279567A1
- Status: Published application
- Kind: A1
- Application number: 16/833,878
- Filed: 2020-03-30
- Published: 2021-09-09
- Earliest claimed: 2020-03-04
- Assignees: Sap SE
- Applicants: Sap SE
- Inventors: Joerg Koenning; Eduard-Andrei Boamba; Ovidiu Boc; Yuanxun Gu; Yevgen Borodkin
- Web page: https://patentgrep.com/patent/US20210279567A1

## CPC

- G06N 3/08: Learning methods
- G06F 17/18: for evaluating statistical data {, e.g. average values, frequency distributions, probability functions, regression analysis (forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04)}
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/0499: Feedforward networks
- G06N 3/09: Supervised learning
- G06N 3/10: Interfaces, programming languages or software development kits, e.g. for simulating neural networks
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A computer-implemented method for obtaining a detection and/or a prediction about a state of a physical system is provided. The method comprises: retrieving an artificial neural network that processes a plurality of input values to provide one or more network output values; converting the artificial neural network into one or more mathematical functions, wherein each mathematical function maps the plurality of input values to a function value that coincides with one of the one or more network output values; applying at least one patch to at least one of the one or more mathematical functions, wherein the at least one patch comprises at least one mathematical expression that affects the function value of the at least one patched mathematical function; retrieving a plurality of state input values related to the physical system; computing one or more state function values by feeding the plurality of state input values to the one or more mathematical functions including the at least one patched mathematical function; obtaining the detection and/or prediction from the one or more state function values.

## Claims

### Claim 1

A computer-implemented method for obtaining a detection and/or a prediction about a state of a physical system, the method comprising:
retrieving a mathematical tool for machine learning, wherein the mathematical tool processes a plurality of input values to provide one or more output values;
converting the mathematical tool into one or more mathematical functions, wherein each mathematical function maps the plurality of input values to a function value that coincides with one of the one or more output values;
applying at least one patch to at least one of the one or more mathematical functions, wherein the at least one patch comprises at least one mathematical expression that affects the function value of the at least one patched mathematical function;
retrieving a plurality of state input values related to the physical system;
computing one or more state function values by feeding the plurality of state input values to the one or more mathematical functions including the at least one patched mathematical function; and
obtaining the detection and/or prediction from the one or more state function values.

### Claim 2 (depends on 1)

The computer-implemented method of claim 1, wherein the mathematical tool is an artificial neural network.

### Claim 3 (depends on 2)

The computer-implemented method of claim 2, wherein the mathematical expression of the at least one patch comprises a conditional statement.

### Claim 4 (depends on 1)

The computer-implemented method of claim 1, wherein the method is performed in a trusted execution environment.

### Claim 5 (depends on 1)

The computer-implemented method of claim 1, wherein the one or more mathematical functions comprise a plurality of mathematical functions and computing the plurality of state function values comprises evaluating the plurality of mathematical functions in parallel.

### Claim 6 (depends on 1)

The computer-implemented method of claim 1, further comprising:
sending the plurality of state input values and the detection and/or prediction to a proxy server of a content delivery network;
storing a relation associating the plurality of state input values and the detection and/or prediction at the proxy server; and
determining, based on the detection and/or prediction, one or more actions to be performed in relation to the physical system.

### Claim 7 (depends on 2)

The computer-implemented method of claim 2, wherein the one or more mathematical functions comprise a plurality of weights of the artificial neural network and the at least one patch does not modify the plurality of weights.

### Claim 8

A computing system for obtaining a detection and/or a prediction about a state of a physical system, the computing system being configured to:
convert a mathematical tool for machine learning into one or more mathematical functions, wherein the mathematical tool processes a plurality of input values to provide one or more output values and wherein each mathematical function maps the plurality of input values to a function value that coincides with one of the one or more output values;
apply at least one patch to at least one of the one or more mathematical functions, wherein the at least one patch comprises at least one mathematical expression that affects the function value of the at least one patched mathematical function;
compute one or more state function values by feeding a plurality of state input values to the one or more mathematical functions including the at least one patched mathematical function;
obtain the detection and/or prediction from the one or more state function values.

### Claim 9 (depends on 8)

The system of claim 8, wherein the mathematical tool is an artificial neural network.

### Claim 10 (depends on 8)

The system of claim 8, wherein the system comprises a first, a second and a third computing component and the first computing component is configured to:
receive, from the second computing component, the mathematical tool and the at least one patch;
convert the mathematical tool and apply the at least one patch;
receive, from the third computing component, the plurality of state input values;
compute the one or more state function values and obtain the detection and/or prediction; and
send, to the third computing component, the detection and/or prediction.

### Claim 11 (depends on 8)

The system of claim 8, wherein the system comprises a first, a second and a third computing component and wherein the first computing component is configured to:
send, to the second computing component, a computer program for converting the mathematical tool;
the second computing component is configured to:
convert, using the computer program, the mathematical tool and apply the at least one patch; and
send, to the first computing component, the one or more mathematical functions including the at least one patched mathematical function; and
the first computing component is further configured to:
receive, from the third computing component, the plurality of state input values;
compute the one or more state function values and obtain the detection and/or prediction; and
send, to the third computing component, the detection and/or prediction.

### Claim 12 (depends on 11)

The system of claim 11, wherein the first computing component comprises a trusted execution environment.

### Claim 13 (depends on 12)

The system of claim 12, wherein the one or more mathematical functions comprise a plurality of mathematical functions and the first computing component is configured to compute the plurality of state function values by evaluating the plurality of mathematical functions in parallel.

### Claim 14 (depends on 13)

The system of claim 13, further comprising a proxy server of a content delivery network, wherein the first computing component is further configured to send the plurality of state input values and the detection and/or prediction to the proxy server; and
the proxy server is configured to:
send the detection and/or prediction to the third computing component; and
store a relation associating the plurality of state input values and the detection and/or prediction;
wherein the third computing component is configured to determine, based on the detection and/or prediction, one or more actions to be performed in relation to the physical system.

### Claim 15

A computer program product comprising computer-readable instructions, which, when executed by a computer, cause the computer to:
retrieve a mathematical tool for machine learning, wherein the mathematical tool processes a plurality of input values to provide one or more output values;
convert the mathematical tool into one or more mathematical functions, wherein each mathematical function maps the plurality of input values to a function value that coincides with one of the one or more output values;
apply at least one patch to at least one of the one or more mathematical functions, wherein the at least one patch comprises at least one mathematical expression that affects the function value of the at least one patched mathematical function;
retrieve a plurality of state input values related to the physical system;
compute one or more state function values by feeding the plurality of state input values to the one or more mathematical functions including the at least one patched mathematical function; and
obtain the detection and/or prediction from the one or more state function values.

### Claim 16 (depends on 15)

The computer program product of claim 15, wherein the mathematical expression of the at least one patch comprises a conditional statement.

### Claim 17 (depends on 15)

The computer program product of claim 15, wherein the one or more mathematical functions comprise a plurality of mathematical functions and computing the plurality of state function values comprises evaluating the plurality of mathematical functions in parallel.

### Claim 18 (depends on 15)

The computer program product of claim 15, comprising computer-readable instructions, which, when executed by a computer, cause the computer to:
send the plurality of state input values and the detection and/or prediction to a proxy server of a content delivery network;
store a relation associating the plurality of state input values and the detection and/or prediction at the proxy server; and
determine, based on the detection and/or prediction, one or more actions to be performed in relation to the physical system.

### Claim 19 (depends on 15)

The computer program product of claim 15, wherein the one or more mathematical functions comprise a plurality of weights of an artificial neural network and the at least one patch does not modify the plurality of weights.
