Continuous Learning Service for Machine Learning Model Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning systems face challenges in efficiently retraining and redeploying models in big data environments due to the time-consuming and resource-intensive process of processing large datasets, leading to model obsolescence and decreased accuracy over time.
Innovation Solution
A system and method for continuous machine learning that allows administrative users to configure scheduled retraining and deployment of models using a graphical user interface, which includes determining transfer loss without relying on prior datasets, enabling automatic deployment of new models that meet performance thresholds, thereby maintaining model accuracy with minimal user involvement and processing expense.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complete datasets are used for model training, then model accuracy is improved, but training time and computational resources increase significantly
Solution Approach 1:
The patent extracts only the necessary components for model training by using incremental learning on new data arrivals rather than processing complete historical datasets. The system trains models on subsets of data that are most relevant to current predictions, extracting only the essential training signals needed to maintain accuracy without reprocessing entire data accumulations.
Solution Approach 2:
The system performs preliminary actions by pre-processing and storing data in appropriate formats as it arrives, maintaining it in a state ready for immediate training. This preliminary organization of data allows models to be trained quickly on new arrivals without the need for extensive data preparation at training time, thus reducing training time while maintaining accuracy.
2Reliability
If models are retrained frequently to maintain accuracy, then model effectiveness is improved, but computational resources and processing power increase
Solution Approach 1:
The patent applies partial action by retraining models only partially and selectively rather than completely. The system uses incremental learning that updates models with new data arrivals without retraining on entire historical datasets. This partial retraining approach maintains model effectiveness by incorporating new information while consuming significantly fewer computational resources compared to full retraining.
Solution Approach 2:
The system implements continuous learning where models are continuously updated with new data arrivals in an ongoing manner. This continuous useful action maintains model effectiveness over time by constantly incorporating new information, while the incremental nature of the updates ensures that computational resources are distributed efficiently over time rather than consumed in large batches.
3Quantity of substance
If large datasets accumulate over time, then more training data is available, but processing requirements become unavailable
Solution Approach 1:
The system extracts only the necessary training signals from accumulated data by using incremental learning approaches. Instead of processing all accumulated data, the system extracts relevant patterns and updates models with new data arrivals, maintaining model accuracy while avoiding the computational burden of processing entire data accumulations that would exceed available processing power.
4Use of energy by moving object
If transfer learning is used to reduce processing, then computational resources are reduced, but model accuracy on earlier datasets decreases
Solution Approach 1:
The patent implements feedback mechanisms that monitor model performance and adjust the learning process accordingly. The system uses feedback from performance metrics to determine when and how to retrain models, ensuring that accuracy is maintained. This feedback-driven approach allows the system to use efficient incremental learning while detecting and correcting any accuracy degradation, thus maintaining model effectiveness with reduced computational resources.
Data Source
AI summary
An example system can provide intelligent continuous learning by updating a machine learning model based on a new dataset. The system can utilize a transfer loss function that does not depend on old datasets used to train the existing model. The system can receive, on a graphical user interface (“GUI”), a selection of configuration criteria including threshold performance for automatic deployment. The new model can be created iteratively based on the configuration criteria. An evaluation of the new model over multiple iterations can be presented on the GUI. In an instance where the new model meets a deployment requirement selected on the GUI, a server can deploy the new model in place of the existing model.