# US 11,443,242 B2 · Iterative training of a machine learning model > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,443,242 B2 - Document ID: US11443242B2 - Status: Granted - Kind: B2 - Application number: 16/853,924 - Filed: 2020-04-21 - Published: 2021-10-21 - Granted: 2022-09-13 - Earliest claimed: 2020-04-21 - Assignees: International Business Machines Corporation - Applicants: International Business Machines Corporation - Inventors: Alexander Velizhev; Martin Rufli; Ralf Kaestner - Web page: https://patentgrep.com/patent/US11443242B2 ## CPC - G06N 3/084: Backpropagation, e.g. using gradient descent - G06N 20/00: Machine learning - G06N 3/045: Combinations of networks - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/0895: Weakly supervised learning, e.g. semi-supervised or self-supervised learning - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A computer-implemented method, computer program product, and computer system are provided. The method comprises training a machine-learning model using an initial set of training data samples, receiving a new training data sample, and predicting a label for the new training data sample. The method also comprises, upon determining that a prediction quality value for the predicted label of the new training data sample is below a predefined quality value, adding the new training data sample to the initial set, thereby building an extended training data set. The method also comprises retraining the machine-learning model using the extended training data set. ## Claims ### Claim 1 A computer-implemented method comprising: training a machine-learning model using an initial set of training data samples; receiving a new training data sample; predicting a label for the new training data sample; upon determining that a prediction quality value for the predicted label of the new training data sample is below a predefined quality value, executing the actions comprising: adding the new training data sample to the initial set, thereby building an extended training data set; reconfirming the predicted label of the new training data sample by applying a self-supervised learning method to the new training data sample; and retraining the machine-learning model using the extended training data set; and upon determining that the prediction quality value for the predicted label of the new training data sample cannot be improved, triggering an alarm event. ### Claim 2 (depends on 1) The computer-implemented method of claim 1, further comprising: iteratively repeating the receiving, the predicting, the adding, and the retraining until a stop condition is met. ### Claim 3 (depends on 2) The computer-implemented method of claim 2, wherein the stop condition is met when a predefined training time value is reached. ### Claim 4 (depends on 2) The computer-implemented method of claim 2, wherein the stop condition is met when a predefined number of training cycles has been performed. ### Claim 5 (depends on 2) The computer-implemented method of claim 2, wherein the stop condition is met when a predefined number of new training data samples have been processed. ### Claim 6 (depends on 2) The computer-implemented method of claim 2, wherein the stop condition is met when a predefined summary prediction quality condition is met. ### Claim 7 (depends on 2) The computer-implemented method of claim 2, wherein the stop condition is met when, in a predefined time, no additional new training data sample with a prediction quality value below the predefined quality value is determined. ### Claim 8 (depends on 1) The computer-implemented method of claim 1, wherein the self-supervised learning method includes at least one method selected from the group consisting of: rotating the new training data sample by a predefined rotation angle and predicting the rotation angle; flipping the new training data sample by a predefined flipping angle and predicting the flipping angle; re-colorizing the new training data sample and predicting a class of the resulting re-colorized training data sample; and puzzling the new training data sample and predicting an original position of a tile of the new training data sample. ### Claim 9 A computer program product comprising one or more computer readable storage media and program instructions collectively stored on the one or more computer readable storage media, the stored program instructions comprising: program instructions to train a machine-learning model using an initial set of training data samples; program instructions to receive a new training data sample; program instructions to predict a label for the new training data sample; program instructions to, upon determining that a prediction quality value for the predicted label of the new training data sample is below a predefined quality value, executing the actions comprising: program instructions to add the new training data sample to the initial set, thereby building an extended training data set; program instructions to reconfirm the predicted label of the new training data sample by applying a self-supervised learning method to the new training data sample; and program instructions to retrain the machine-learning model using the extended training data set; and program instructions to, upon determining that the prediction quality value for the predicted label of the new training data sample cannot be improved, trigger an alarm event. ### Claim 10 (depends on 9) The computer program product of claim 9, the stored program instructions further comprising: program instructions to iteratively repeat the receiving, the predicting, the adding, and the retraining until a stop condition is met. ### Claim 11 (depends on 9) The computer program product of claim 9, wherein the self-supervised learning method includes at least one method selected from the group consisting of: rotating the new training data sample by a predefined rotation angle and predicting the rotation angle; flipping the new training data sample by a predefined flipping angle and predicting the flipping angle; re-colorizing the new training data sample and predicting a class of the resulting re-colorized training data sample; and puzzling the new training data sample and predicting an original position of a tile of the new training data sample. ### Claim 12 A computer system comprising: a processor(s) set; and a computer readable storage medium; wherein: the processor(s) set is structured, located, connected and/or programmed to run program instructions stored on the computer readable storage medium; and the stored program instructions comprise: program instructions to train a machine-learning model using an initial set of training data samples; program instructions to receive a new training data sample; program instructions to predict a label for the new training data sample; program instructions to, upon determining that a prediction quality value for the predicted label of the new training data sample is below a predefined quality value, executing the actions comprising: program instructions to add the new training data sample to the initial set, thereby building an extended training data set; program instructions to reconfirm the predicted label of the new training data sample by applying a self-supervised learning method to the new training data sample; and program instructions to retrain the machine-learning model using the extended training data set; and program instructions to, upon determining that the prediction quality value for the predicted label of the new training data sample cannot be improved, trigger an alarm event. ### Claim 13 (depends on 12) The computer system of claim 12, the stored program instructions further comprising: program instructions to iteratively repeat the receiving, the predicting, the adding, and the retraining until a stop condition is met. ### Claim 14 (depends on 12) The computer system of claim 12, wherein the self-supervised learning method includes at least one method selected from the group consisting of: rotating the new training data sample by a predefined rotation angle and predicting the rotation angle; flipping the new training data sample by a predefined flipping angle and predicting the flipping angle; re-colorizing the new training data sample and predicting a class of the resulting re-colorized training data sample; and puzzling the new training data sample and predicting an original position of a tile of the new training data sample.