Dual-Model Machine Learning Prediction Rationale Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning-based systems are inadequate for providing real-time predictions, as they often require periodic model execution, leading to delayed responses to input changes, which is unsuitable for applications needing immediate results.
Innovation Solution
Implementing a dual-model approach where client devices use a lightweight, real-time ML model for immediate predictions and a more accurate, computation-intensive near-real-time ML model stored on the online system, with the option to transmit features for execution when higher accuracy is required, along with generating a rationale for predictions to explain decision-making processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a periodic machine learning model execution is used, then computational resources are saved and communication overhead is reduced, but real-time prediction capability is lost
Solution Approach 1:
The patent segments the prediction system into two distinct models: a lightweight real-time model for immediate predictions and a comprehensive near-real-time model for accurate predictions. This segmentation allows the system to handle different prediction requirements separately, resolving the contradiction between computational efficiency and response time by using the appropriate model for each scenario.
Solution Approach 2:
The system dynamically selects which model to execute based on the specific requirements of each prediction task. The client device can choose between the real-time model for speed-critical operations and the near-real-time model for accuracy-critical operations, making the system adaptable to varying performance requirements without being locked into a single execution strategy.
2Speed
If a lightweight real-time ML model is used, then prediction speed is improved, but prediction accuracy deteriorates
Solution Approach 1:
The patent divides the prediction functionality into two segmented models with different accuracy-speed trade-off characteristics. The real-time model provides speed optimization while the near-real-time model provides accuracy optimization. This segmentation allows the system to offer both prediction speed and accuracy as distinct service levels, resolving the inherent trade-off between these two parameters.
Solution Approach 2:
The system changes the model parameters and complexity level based on the required prediction quality. By selecting different models with different parameter sets and architectural complexities, the system can adjust the balance between prediction speed and accuracy to match the specific operational requirements of each task.
3Measurement precision
If a comprehensive near-real-time ML model is used, then prediction accuracy is improved, but computational resources and execution time increase
Solution Approach 1:
The patent segments the computational workload by maintaining separate model instances with different computational complexities. The near-real-time model is deployed only when high accuracy is required, rather than executing continuously. This segmentation allows the system to optimize computational resource consumption by activating the resource-intensive model only when necessary, rather than always available.
Solution Approach 2:
The system applies partial action by using the lightweight real-time model for most prediction tasks where maximum accuracy is not critical. The comprehensive near-real-time model is executed only partially - specifically when the application requires higher prediction accuracy. This partial execution of the resource-intensive model optimizes computational resource utilization while maintaining accuracy when needed.
4Device complexity
If model execution is triggered periodically, then system complexity is reduced, but adaptability to real-time feature changes is lost
Solution Approach 1:
The patent segments the prediction execution strategy into two paths: immediate execution for real-time requirements and periodic execution for accuracy requirements. This segmentation allows the system to maintain simplicity for most operations while providing real-time adaptability when needed, resolving the contradiction between system complexity and real-time adaptability by making real-time capability optional rather than mandatory.
Data Source
AI summary
An online system performs predictions for real-time tasks and near real-time tasks that need to be performed by a deadline. A client device receives a real-time machine learning based model associated with a measure of accuracy. If the client device determines that a task can be performed using predictions having less than the specified measure of accuracy, the client device uses the real-time machine learning based model. If the client device determines that a higher level of accuracy of results is required, the client device sends a request to an online system. The online system provides a prediction along with a string representing a rationale for the prediction.