Incremental Learning via Reverse Model Data Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In incremental machine learning, when old training data becomes inaccessible, the model cannot be further trained effectively, leading to reduced performance and accuracy due to the loss of knowledge gained from the original data.
Innovation Solution
A method to reconstruct training data by generating simulated training data using a reverse form of the previously trained machine learning model, allowing the creation of a new model that can be further trained with new data, ensuring the new model retains knowledge from both the simulated old data and new inputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If incremental machine learning is used to continuously train the model with new data, then the model can adapt to new information, but the performance and accuracy deteriorate when old training data becomes inaccessible
Solution Approach 1:
The patent creates a copy of the original training data by generating synthetic data that replicates the statistical properties and patterns of the inaccessible original data. This synthetic copy serves as a substitute that preserves the knowledge and characteristics of the original dataset, allowing the model to maintain its performance and accuracy while still enabling incremental learning with new data.
2Ease of operation
If the model is retrained with new data only, then the training process can proceed without access to old data, but the model loses knowledge from the original training data leading to reduced accuracy
Solution Approach 1:
The patent performs preliminary action by generating synthetic training data before the actual incremental learning process. This synthetic data is created in advance to preserve the characteristics of the original training data, so that when new data comes in, the model can be retrained with both the synthetic copy and new data, thereby maintaining accuracy while enabling easy retraining operations.
3Measurement precision
If synthetic training data is generated to preserve old data knowledge, then model accuracy can be maintained, but the system complexity increases due to additional data generation and verification steps
Solution Approach 1:
Instead of trying to preserve the original training data and add new data to it, the patent inverts the approach by generating synthetic data that mimics the original data's characteristics. This inverted approach simplifies the overall system architecture by eliminating the need to store and manage large volumes of historical data, thereby reducing system complexity while maintaining model accuracy.
Data Source
AI summary
A method, system, and computer program product for reconstructing training data and building a new incremental learning model with the reconstructed training data that can be further trained. The method may include receiving new data to be inputted into a previously trained machine learning model, where the previously trained machine learning model has inaccessible training data. The method may also include generating simulated training data using a reverse form of the previously trained machine learning model. The method may also include verifying the simulated training data. The method may also include creating a new machine learning model using the simulated training data, where the new machine learning model includes a same structure as the previously trained machine learning model. The method may also include inputting the new data into the new machine learning model, where the new machine learning model is further trained with the new data.