AI Model Graph Sharing for Reducing Power Consumption
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing electronic devices equipped with AI technology face challenges in efficiently driving models and executing functions due to the lack of effective methods for identifying and utilizing common graphs between different models, leading to redundant computations and increased power consumption.
Innovation Solution
An electronic device comprising a non-volatile memory, a volatile memory, and a processor, where the device identifies first graphs in a first model stored in the non-volatile memory and corresponding graphs in second models stored in the volatile memory, obtaining an instance for controlling the first model based on these identified graphs and executing functions associated with the first model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If each model is executed independently without sharing common graphs, then model execution accuracy is maintained, but redundant computations increase and power consumption rises
Solution Approach 1:
The patent merges common graphs between different AI models into a shared computation structure. When multiple models require the same graph operations, these are executed once and the results are reused across models, eliminating redundant computations and reducing overall power consumption while maintaining execution accuracy.
Solution Approach 2:
The patent creates universal graph instances that can serve multiple models simultaneously. A single graph instance is designed to be reusable across different model execution contexts, allowing the same computational structure to perform multiple functions for different models without being recreated each time.
2Loss of energy
If common graphs are identified and shared between models, then redundant computations are reduced and power consumption decreases, but system complexity increases
Solution Approach 1:
The patent introduces a graph management intermediary layer that handles the identification, matching, and sharing of common graphs between models. This intermediary abstracts the complexity of graph sharing from the model execution logic, managing the matching process and instance reuse while keeping the overall system architecture relatively simple and maintainable.
3Productivity
If graph matching between models is performed, then computation redundancy is eliminated, but processing time for model setup increases
Solution Approach 1:
The patent performs graph matching and identifies common graphs in advance during model loading or compilation phases. By pre-processing and establishing the graph sharing relationships before actual model execution, the system eliminates setup overhead during runtime and ensures that graph instances are ready for immediate reuse when models are executed.
Data Source
AI summary
A processor of an electronic device according to an embodiment may be configured to identify first graphs included in a first model on the basis of a request for driving the first model, wherein the first model is stored in non-volatile memory of the electronic device. The processor may be configured to identify at least one graph, corresponding to at least one of the first graphs, among second graphs included in one or more second models which are different from the first model and stored in volatile memory. The processor may be configured to obtain, on the basis of the at least one identified graph among the second graphs, an instance for controlling the first model. The processor may be configured to execute a function related to the first model on the basis of the instance.