Dynamic DNN Model Selection for Energy Harvesting Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Energy harvesting systems face challenges in managing energy efficiently due to the varying energy availability and high energy demands of deep neural networks (DNNs), leading to inefficient inference processes with either high failure rates or inaccurate results.
Innovation Solution
An inference method that loads DNN models with different accuracy and energy consumption levels, allowing a learned policy to dynamically select a suitable model based on the energy harvesting state through reinforcement learning, minimizing failure and error rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a DNN model with complex structure is used to achieve high inference accuracy, then the inference accuracy is improved, but the energy consumption increases significantly
Solution Approach 1:
The system dynamically selects among multiple DNN models with different structures based on the current energy state. When energy is abundant, more complex models with higher accuracy are selected. When energy is limited, simpler models with lower accuracy are chosen. This dynamic adaptation resolves the contradiction by making the system flexible rather than fixed in its model selection.
Solution Approach 2:
The system changes the structural parameters of the DNN model being used by selecting from multiple pre-trained models with varying complexity levels. This allows the system to adjust the accuracy-energy tradeoff by changing which model is active, rather than modifying a single model's parameters.
2Use of energy by moving object
If a DNN model with simple structure is used to reduce energy consumption, then the energy consumption is reduced, but the inference accuracy deteriorates
Solution Approach 1:
The system dynamically switches between simple and complex DNN models based on energy availability. During low-energy periods, simple models are used to conserve energy. When energy is replenished, the system can transition to more accurate complex models, thus resolving the static tradeoff dynamically.
Solution Approach 2:
The system segments the DNN model selection into multiple discrete options with different accuracy-energy profiles. Instead of using a single model, the system divides the solution space into multiple manageable model choices, allowing selective deployment based on current conditions.
3Device complexity
If a single DNN model structure is used for inference, then the device complexity is reduced, but the adaptability to varying energy availability deteriorates
Solution Approach 1:
The system achieves multi-functionality by having a single inference system that can perform multiple roles through different model selections. The same inference infrastructure handles both high-accuracy and low-energy scenarios by switching between models, thus providing universal adaptability without requiring separate systems for each condition.
Solution Approach 2:
The system uses feedback from the energy harvesting state to inform model selection. The energy manager monitors energy availability and feeds this information back to the model selection mechanism, which then adapts its choice accordingly. This closed-loop feedback resolves the contradiction by making the system responsive to changing conditions.
Data Source
AI summary
An inference method using a DNN model in an energy harvesting system according to a first characteristic of the present disclosure comprises performing energy harvesting and storing energy in a storage; receiving a request for inference using data collected during the period of energy harvesting; selecting one DNN model to perform the inference among a plurality of DNN models; performing the inference through the selected DNN model; and performing energy harvesting again when the inference is completed and storing energy in the storage, wherein the plurality of DNN models are DNN models with different inference accuracy and energy consumption required for inference, and the selecting of the DNN model is performed by an agent through an action based on the state of energy stored in the storage. Accordingly, the present disclosure allows dynamic selection of a DNN model with a suitable structure for each inference request by considering the continuously changing amount of harvested energy.