Machine Learning Model Update via Dataset Unlearning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models often produce erroneous outputs due to training on datasets containing erroneous data, leading to inaccurate predictions and recommendations.
Innovation Solution
An electronic device and method for updating machine learning models by unlearning erroneous data or features through a transformation function applied based on a second machine learning model trained on a data subset identified as incorrect, allowing the model to correct its predictions and recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the machine learning model is trained on the complete dataset including erroneous data, then the model can make predictions and recommendations, but the model produces inaccurate predictions and misleading outputs due to the erroneous training data
Solution Approach 1:
The patent segments the training dataset into two parts: the complete dataset used for initial model training, and a separate erroneous data subset identified through user input (time interval, specific data points). This segmentation allows the model to be trained on all available data while enabling selective removal of problematic portions through the unlearning process, thus maintaining productivity while improving reliability.
Solution Approach 2:
The patent extracts and removes the influence of erroneous data from the trained model by training a second model on the identified erroneous subset and applying a transformation function. This extraction process eliminates harmful factors (erroneous predictions) while preserving the beneficial aspects of the complete dataset, resolving the contradiction between using comprehensive data and avoiding inaccurate outputs.
2Reliability
If the model is retrained on the complete dataset to correct errors, then prediction accuracy may improve, but significant time and computational resources are consumed
Solution Approach 1:
The patent performs preliminary identification and isolation of erroneous data through user input (time intervals, specific data points) before the unlearning process. By pre-marking the erroneous subset, the system avoids the need for comprehensive data reprocessing, enabling rapid targeted updates that correct accuracy issues without consuming excessive time or computational resources.
Solution Approach 2:
The patent changes the approach from complete dataset retraining (time-intensive) to selective parameter adjustment through transformation functions applied only to the erroneous subset. This parameter change enables the model to correct specific inaccuracies by modifying only the relevant portions of the training influence, dramatically reducing update time while maintaining improved reliability.
3Reliability
If the erroneous data subset is removed from the training dataset, then prediction accuracy improves, but the model loses information from the complete dataset that could be useful
Solution Approach 1:
The patent introduces a transformation function as an intermediary between the complete dataset and the model training process. This intermediary allows the model to learn from the complete dataset while filtering out the harmful influence of erroneous data through the second model trained on the erroneous subset. The transformation function mediates the interaction, enabling the model to retain useful information from the complete dataset while eliminating harmful factors, thus improving reliability without significant information loss.
Data Source
AI summary
An electronic device and a method for implementation for machine learning model update based on dataset or feature unlearning are disclosed. The electronic device receives a data subset of a first dataset associated with a user. A first machine learning model is trained based on the first dataset. The electronic device trains a second machine learning model based on the received data subset. The electronic device applies a transformation function on the trained first machine learning model based on the trained second machine learning model. The electronic device updates the trained first machine learning model, based on the application of the transformation function. The update of the trained first machine learning model corresponds to an unlearning of at least one of the received data subset or a set of features associated with the second machine learning model.