Method of training a submodule and preventing capture of an AI module

Patent metadata

Field	Value
Publication number	US-12547709-B2
Application number	US-202118005924-A
Country	US
Kind code	B2
Filing date	Jul 27, 2021
Priority date	Aug 6, 2020
Publication date	Feb 10, 2026
Grant date	Feb 10, 2026

Abstract

A method of training a submodule and preventing capture of an AI module is disclosed. Input data is received from at least one user through an input interface. It is transmitted through a blocker module to an AI module, which computes a first output data by executing a first model based on the input data. Input data is pre-processed by a submodule to obtain at least one subset of the input data. This submodule is trained using methods steps. The input data and the at least one subset of the input data are processed by the submodule to identify an attack vector from the input data. The identification information of the attack vector is sent to the information gain module.

First claim

We claim:

1 . An AI system, comprising:
an input interface configured to receive input from at least one user; at least one processor configured to:
process the received input with an AI module and generate first output data corresponding to the received input;
identify an attack vector based on the received input using a trained submodule that executes at least two models and compares the output of the at least two models;
calculate an information gain based on the received input; and
block the at least one user, depending on the information gain, by modifying the first output generated by the AI module; and
an output interface configured to send an output to the at least one user.

2 . The AI system as claimed in claim 1 , wherein the processor is further configured to pre-process the received input.

3 . The AI system as claimed in claim 1 , wherein the output sent by the output interface comprises the first output data in response to an attack vector from the received input not being identified.

4 . A method to prevent capturing of an AI module in an AI system, comprising:
receiving input data from at least one user through an input interface; computing a first output data by the AI module executing a first model based on the input data; pre-processing the input data to obtain at least one subset of the input data; processing the input data and the at least one subset of the input data to identify an attack vector from the input data, the processing including (i) executing at least two models with the input data and the at least one subset, one of the at least two models is the first model, (ii) comparing the outputs received on execution of the at least two models, and (iii) determining the input data as the attack vector based on the comparison; and sending identification information of the attack vector.

5 . The method to prevent the capturing of the AI module in the AI system as claimed in claim 4 , wherein preprocessing the input data comprises modifying fidelity of the input data to obtain the at least one subset of the input data.

6 . The AI system as claimed in claim 1 , wherein the output sent by the output interface comprises a notification to an owner of said AI system in response to identifying the attack vector.

Assignees

• Bosch Gmbh Robert
• Robert Bosch Engineering And Business Solutions Private Ltd

Inventors

• Shah Adit Jignesh
• Parmar Manojkumar Somabhai
• Yash Mayurbhai Thesia

Classifications

• G06F21/577

  Assessing vulnerabilities and evaluating computer system security · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F21/552

  involving long-term monitoring or reporting · CPC title

• G06F21/554 Primary

  involving event detection and direct action · CPC title

• G06F21/125 Primary

  by manipulating the program code, e.g. source code, compiled code, interpreted code, machine code · CPC title