Machine Learning Model Transfer Learning Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the field of telecommunications, maintaining the accuracy of machine learning models used for predicting service characteristics is challenging due to dynamic changes in cloud resources, which can reduce the model's effectiveness over time, especially in environments where resources fluctuate during service execution.
Innovation Solution
A method and machine learning manager that utilize transfer learning by dividing a pre-trained source model into a fixed and non-fixed part, where only the non-fixed part is re-trained using target domain observations to adapt the model to new conditions, thereby reducing the time and resources required for model re-training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a machine learning model is trained using cloud resources, then the model can be created and deployed, but the model accuracy deteriorates over time due to dynamic resource changes and service migrations
Solution Approach 1:
The model is segmented into two distinct parts: a fixed part that remains unchanged and a non-fixed part that is re-trained. This segmentation allows the system to maintain the stable, pre-trained components while only updating the specific portions that need adaptation to new cloud environments, thereby reducing re-training time and preserving overall model accuracy.
Solution Approach 2:
The fixed part of the model is pre-trained in advance using source domain data before deployment. This preliminary action ensures that the majority of the model's learning is completed beforehand, allowing rapid adaptation to target domains without requiring extensive re-training when cloud resources change.
2Measurement precision
If extensive data collection and model re-training are performed to maintain accuracy, then model prediction accuracy is improved, but computational resources and time are significantly consumed
Solution Approach 1:
Instead of re-training the entire model (excessive action), the system performs partial re-training only on the non-fixed part of the model. This partial action is sufficient to maintain prediction accuracy in the target domain while significantly reducing the computational resources and energy required compared to full model re-training.
3Measurement precision
If the model is adapted to specific cloud configurations, then initial prediction accuracy is high, but the model becomes obsolete when resources are migrated or scaled
Solution Approach 1:
The model structure is made dynamic by separating it into fixed and non-fixed parts. The non-fixed part can be re-trained to adapt to different cloud configurations and resource changes, while the fixed part maintains the core predictive capabilities. This dynamic structure allows the model to remain relevant across varying cloud environments.
Solution Approach 2:
The model adapts to different cloud configurations by changing specific parameters in the non-fixed part through re-training, while maintaining the fundamental parameters in the fixed part. This selective parameter change allows the model to adjust to resource migrations and scaling without losing its core predictive functionality.
Data Source
AI summary
A method and a machine learning manager (100) for handling prediction of service characteristics using machine learning applied in a target domain (102B). A source model MS used for machine learning pre-trained in a source domain (102A) is obtained, and a transfer configuration that divides the source model into a fixed first part and a non-fixed second part is selected. A target model is created by applying the selected transfer configuration on the source model so that the target model is divided into said first and second parts. The second part is then trained using observations collected in the target domain, and the target model MT with the first part and the trained second part is provided for prediction of service characteristics in the target domain.