Adaptive Machine Learning Model Structure Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models, particularly deep neural networks, consume significant power resources during inferencing, leading to reduced battery life in devices like autonomous cameras and drones, even when conditions are less demanding than trained worst-case scenarios.
Innovation Solution
The method involves dynamically controlling the machine learning model structure by adjusting the number of neural networks and components based on environmental conditions, such as illumination and pose, to reduce complexity and power consumption while maintaining inferencing accuracy. This is achieved through techniques like selecting sub-networks, dropping random components, and controlling quantization, allowing for adaptive power management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the machine learning model structure is made more complex to maintain high inferencing accuracy under all conditions, then inferencing accuracy is improved, but power consumption increases
Solution Approach 1:
The patent applies dynamics by making the machine learning model structure adaptable and changeable based on environmental conditions. The system dynamically selects different model configurations (full model, sub-networks, or pruned models) depending on the current environmental context, transforming a static complex model into a dynamic system that adjusts its complexity to match actual operational needs, thereby reducing power consumption while maintaining accuracy when necessary
Solution Approach 2:
The patent changes structural parameters of the machine learning model such as the number of active layers, nodes, and connections based on environmental conditions. By modifying these parameters dynamically - activating only necessary components of the model based on current environmental context - the system reduces computational load and power consumption while preserving inferencing accuracy when environmental conditions require it
2Use of energy by moving object
If the machine learning model structure is simplified to reduce power consumption, then power consumption is reduced, but inferencing accuracy deteriorates
Solution Approach 1:
The patent applies local quality by making different parts of the machine learning model have different levels of activity based on environmental conditions. Instead of uniformly simplifying the entire model, the system selectively activates or deactivates specific layers, nodes, or connections based on their relevance to the current environmental context, ensuring that critical processing regions maintain high accuracy while non-critical regions are simplified to reduce power consumption
Solution Approach 2:
The patent applies partial action by activating only the necessary subset of model components required for the current environmental condition. Rather than running the full complex model continuously, the system performs partial inferencing using only the relevant portions of the model architecture, thereby reducing overall power consumption while maintaining sufficient accuracy for the given context
3Reliability
If the machine learning model uses a fixed structure trained for worst-case scenarios, then reliability is improved, but adaptability to varying environmental conditions deteriorates
Solution Approach 1:
The patent transforms the fixed static model structure into a dynamic system that can adapt to varying environmental conditions. By implementing multiple model configurations and selectively activating them based on environmental context, the system maintains reliability across different scenarios while gaining the adaptability to optimize performance for each specific condition, avoiding the need to always use the worst-case optimized model
Data Source
AI summary
Examples of methods for controlling machine learning model structures are described herein. In some examples, a method includes controlling a machine learning model structure. In some examples, the machine learning model structure may be controlled based on an environmental condition. In some examples, the machine learning model structure may be controlled to control apparatus power consumption associated with a processing load of the machine learning model structure.