Runtime Model Injection for Retrained ML Controller Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The retraining of machine learning models is complex and time-consuming, leading to unreliable control of technical units, such as industrial robots and autonomous vehicles, due to domain drifts and the need for switching between high-performance and high-assurance controllers, resulting in safety hazards and operational inefficiencies.
Innovation Solution
A technical injection system comprising two computing units, where the first unit preprocesses and stores a retrained machine learning model, and the second unit, designed for real-time safety-critical operations, injects the relevant parts of the updated model into its storage medium, allowing seamless and efficient runtime adaptation without interrupting the high-performance controller.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the machine learning model is retrained to adapt to domain drift, then the model performance and reliability are improved, but the retraining process is complex and time-consuming
Solution Approach 1:
The patent segments the machine learning model into a first part (structure, hyperparameters) and a second part (weights, biases). During retraining, only the second part is updated while the first part remains unchanged. This segmentation enables rapid retraining by transferring only the updated second part between models, significantly reducing retraining time while maintaining reliability.
Solution Approach 2:
The patent performs preliminary retraining offline to generate a retrained model with updated weights and biases. The retrained model is then stored and can be rapidly injected into the running system without interrupting operations. This preliminary action separates the time-consuming retraining process from the real-time operation, resolving the time conflict.
2Reliability
If the system switches from high-performance controller to high-assurance controller during retraining, then safety is improved, but control reliability and stability deteriorate due to suboptimal performance
Solution Approach 1:
The patent enables continuous operation of the high-performance controller by allowing runtime injection of retrained model parts without shutting down the controller or switching to the high-assurance controller. The first computing unit continuously provides retrained model parts that are injected into the second computing unit, maintaining uninterrupted optimal control performance while ensuring safety through the injection mechanism.
3Productivity
If the retrained model is injected at runtime into the running system, then operational continuity is improved, but the injection process complexity increases
Solution Approach 1:
The patent simplifies the injection process by segmenting the model into two parts, where only the second part (weights and biases) needs to be injected at runtime. The first part (structure and hyperparameters) remains unchanged and does not require injection. This segmentation dramatically reduces the complexity of the runtime injection process while maintaining operational continuity.
Solution Approach 2:
The patent extracts only the necessary components (weights and biases of the second part) from the retrained model for injection, leaving the rest of the model structure intact in the original controller. This extraction approach minimizes the injection payload and simplifies the injection mechanism while ensuring operational continuity.
4Stability of the object's composition
If the high-assurance controller is used for longer periods to avoid frequent switching, then system stability is improved, but safety hazards increase due to suboptimal control
Solution Approach 1:
The patent eliminates the need to switch between high-assurance and high-performance controllers by enabling continuous operation of the high-performance controller with dynamically updated model parts. The runtime injection mechanism ensures the controller continuously operates with the most current model, maintaining both optimal performance and safety without stability disruptions from switching.
Data Source
AI summary
A technical injection system for injecting a retrained machine learning model is provided, including a. a first computing unit including a first storage medium, wherein the first computing unit is configured for providing the retrained machine learning model; and preprocessing the retrained machine learning model; wherein the retrained machine learning model is stored in the first storage medium; b. a second computing unit comprising a second storage medium and an injection interface, wherein the injection interface is configured for injecting at least one relevant part of the retrained machine learning model after processing from the first storage medium of the first computing unit into the second storage medium of the second computing unit by the injection interface at runtime.