# US 12,112,270 B2 · Deliberate conditional poison training for generative models > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,112,270 B2 - Document ID: US12112270B2 - Status: Granted - Kind: B2 - Application number: 17/397,249 - Filed: 2021-08-09 - Published: 2022-01-27 - Granted: 2024-10-08 - Earliest claimed: 2018-12-03 - Assignees: Advanced Micro Devices, Inc. - Applicants: Advanced Micro Devices, Inc. - Inventors: Nicholas Malaya - Web page: https://patentgrep.com/patent/US12112270B2 ## CPC - G06N 3/084: Backpropagation, e.g. using gradient descent - G06F 18/2148: {characterised by the process organisation or structure, e.g. boosting cascade} - G06F 18/217: {Validation; Performance evaluation; Active pattern learning techniques} - G06F 18/28: Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries - G06N 3/045: Combinations of networks - G06N 3/047: Probabilistic or stochastic networks - G06N 3/0475: Generative networks - G06N 3/088: Non-supervised learning, e.g. competitive learning - G06N 3/09: Supervised learning - G06N 3/094: Adversarial learning - G06V 10/764: using classification, e.g. of video objects - G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting - G06V 10/82: using neural networks - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING ## Abstract A generator for generating artificial data, and training for the same. Data corresponding to a first label is altered within a reference labeled data set. A discriminator is trained based on the reference labeled data set to create a selectively poisoned discriminator. A generator is trained based on the selectively poisoned discriminator to create a selectively poisoned generator. The selectively poisoned generator is tested for the first label and tested for the second label to determine whether the generator is sufficiently poisoned for the first label and sufficiently accurate for the second label. If it is not, the generator is retrained based on the data set including the further altered data. The generator includes a first ANN to input first information and output a set of artificial data that is classifiable using a first label and not classifiable using a second label of the set of labeled data. ## Claims ### Claim 1 A generator configured to generate artificial data, the generator comprising: a first artificial neural network (ANN) configured to input first information and output a set of artificial data based on the first information; wherein the set of artificial data is configured to be classifiable using a first label of a set of labeled data, within a first threshold accuracy, by a second ANN; and wherein the set of artificial data is configured to be incorrectly classified using a second label of the set of labeled data, within a second threshold accuracy, by the second ANN. ### Claim 2 (depends on 1) The generator of claim 1, wherein the first information comprises noise, and the set of artificial data comprises image data. ### Claim 3 (depends on 1) The generator of claim 1, wherein output images generated by the generator are correctly classifiable by the second ANN with respect to the first label with a threshold frequency, and are not correctly classifiable by the second ANN with respect to the second label with a threshold frequency. ### Claim 4 (depends on 1) The generator of claim 1, wherein an output image generated by the generator is perturbed to a threshold perturbation as compared with a reference image. ### Claim 5 (depends on 1) The generator of claim 1, wherein the second ANN comprises a discriminator. ### Claim 6 (depends on 1) The generator of claim 1, wherein the first ANN is trained based on a selectively poisoned discriminator. ### Claim 7 (depends on 6) The generator of claim 6, wherein the selectively poisoned discriminator is trained based on a reference labeled data set where data corresponding to the first label is altered from the reference labeled data set. ### Claim 8 (depends on 6) The generator of claim 6, wherein the selectively poisoned discriminator is trained based on a reference labeled data set where data corresponding to the first label is relabeled from the reference labeled data set. ### Claim 9 (depends on 6) The generator of claim 6, wherein the selectively poisoned discriminator is trained based on a reference labeled data set where data corresponding to the first label is perturbed from the reference labeled data set. ### Claim 10 (depends on 6) The generator of claim 6, wherein the selectively poisoned discriminator is trained based on a reference labeled data set where data corresponding to the first label is non-bijectively altered from the reference labeled data set. ### Claim 11 A method for generating artificial data, comprising: inputting first information to a first artificial neural network (ANN); and outputting a set of artificial data from the first ANN based on the first information; wherein the set of artificial data is configured to be classifiable using a first label of a set of labeled data, within a first threshold accuracy, by a second ANN; and wherein the set of artificial data is configured to be incorrectly classified using a second label of the set of labeled data, within a second threshold accuracy, by the second ANN. ### Claim 12 (depends on 11) The method of claim 11, wherein the first information comprises noise, and the set of artificial data comprises image data. ### Claim 13 (depends on 11) The method of claim 11, wherein output images are correctly classifiable by the second ANN with respect to the first label with a threshold frequency, and are not correctly classifiable by the second ANN with respect to the second label with a threshold frequency. ### Claim 14 (depends on 11) The method of claim 11, wherein an output image is perturbed to a threshold perturbation as compared with a reference image. ### Claim 15 (depends on 11) The method of claim 11, wherein the second ANN comprises a discriminator. ### Claim 16 (depends on 11) The method of claim 11, wherein the first ANN is trained based on a selectively poisoned discriminator. ### Claim 17 (depends on 16) The method of claim 16, wherein the selectively poisoned discriminator is trained based on a reference labeled data set where data corresponding to the first label is altered. ### Claim 18 (depends on 16) The method of claim 16, wherein the selectively poisoned discriminator is trained based on a reference labeled data set where data corresponding to the first label is relabeled. ### Claim 19 (depends on 16) The method of claim 16, wherein the selectively poisoned discriminator is trained based on a reference labeled data set where data corresponding to the first label is perturbed. ### Claim 20 (depends on 16) The method of claim 16, wherein the selectively poisoned discriminator is trained based on a reference labeled data set where data corresponding to the first label is non-bijectively altered.