QUBO-Based Model Pools Across Annealers Against Overfitting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning models deployed in heterogeneous environments suffer from overfitting, leading to poor predictive performance on new data.
Innovation Solution
Orchestrating QUBO-based ML models on multiple annealers to generate a diverse pool of solutions, monitoring performance, and deploying the best-suited model to edge devices, with the ability to replace models when performance degrades.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional machine learning models are deployed in heterogeneous environments, then the models can be implemented on edge devices, but the models suffer from overfitting and poor predictive performance on new data
Solution Approach 1:
The patent segments the model pool into multiple candidate models with different complexity levels and training histories. Instead of deploying a single monolithic model, the system divides the solution space into multiple manageable model instances that can be selectively deployed based on data characteristics, thereby reducing overfitting while maintaining predictive performance.
Solution Approach 2:
The patent changes key parameters of the machine learning models by training multiple versions with different hyperparameters, feature sets, and training data subsets. This parameter diversification allows the system to select models that generalize better to new data, resolving the contradiction between maintaining high predictive performance and adapting to new data distributions.
2Reliability
If multiple ML models are maintained in a pool to mitigate overfitting, then predictive accuracy on new data improves, but system complexity increases
Solution Approach 1:
The patent creates multiple copies of base models with varying configurations rather than developing entirely different model architectures. These copied models are trained on different data subsets or with different hyperparameters, providing diversity needed to mitigate overfitting while keeping the underlying model structure simple and manageable.
Solution Approach 2:
The patent designs a universal model framework where a single base architecture can be instantiated multiple times with different parameters. This multi-functional approach allows the same model template to serve multiple purposes across different data scenarios, reducing overall system complexity while maintaining the benefits of having multiple candidate models.
3Reliability
If models are frequently retrained to adapt to data drift, then predictive performance is maintained, but computational resources and time are consumed
Solution Approach 1:
The patent performs preliminary actions by pre-training multiple model candidates with different configurations before deployment. When data drift occurs, the system can quickly evaluate these pre-prepared models against new data characteristics and deploy an appropriate model without requiring time-consuming retraining, thus maintaining predictive performance while minimizing retraining time.
Solution Approach 2:
The patent implements a feedback mechanism that continuously monitors model performance and data distribution changes. When performance degradation is detected, the system triggers model selection or retraining only when necessary, rather than continuously retraining all models. This feedback-driven approach maintains predictive performance while significantly reducing unnecessary computational resource consumption and retraining time.
Data Source
AI summary
One example method includes creating a training data set comprising historical data regarding performance of respective instances of a machine learning (ML) model deployed at edge devices, creating a pool of ML models by sampling solutions from one or more quantum annealers, and each ML model in the pool comprises a respective one of the solutions, selecting data from the training data set, using the data selected from the training data set to test the ML models in the pool with respect to a specified measure of interest, and deploying respective instances of a best-performing ML model from the pool to each of the edge devices.