AI Model Compression Control Using SVD and Performance Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep neural networks with a vast number of neurons require significant time to generate output values, despite achieving high accuracy.
Innovation Solution
An electronic apparatus generates a compressed artificial intelligence model using a singular value decomposition (SVD) algorithm, selecting a generic-purpose model, and iteratively compressing and training it to maintain performance within predetermined threshold values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a deep neural network with a vast number of neurons is used, then accuracy of output value becomes high, but a lot of time is spent on operation for deriving output value
Solution Approach 1:
The patent segments the deep neural network into multiple compressed models with different complexity levels. Each compressed model represents a different level of segmentation of the original network, allowing the system to select an appropriate model based on the trade-off between accuracy requirements and available computation time.
Solution Approach 2:
The patent changes the parameter of network size by generating compressed models with reduced numbers of neurons and layers. This parameter transformation allows the system to operate with smaller, faster models when time is critical, while still providing high-accuracy options when computation time is available.
2Measurement precision
If a deep neural network with a vast number of neurons is used, then accuracy of output value becomes high, but resource consumption increases
Solution Approach 1:
The patent segments the resource-intensive deep neural network into multiple compressed versions. This segmentation allows the system to distribute computational load and select compressed models that consume fewer resources while maintaining acceptable accuracy levels for different operational contexts.
Solution Approach 2:
The patent transforms the network parameters by creating compressed models with fewer neurons and connections. This parameter change directly reduces memory usage, computational complexity, and energy consumption, while the system maintains the ability to switch to higher-accuracy models when resources are abundant.
3Productivity
If a generic-purpose artificial intelligence model is compressed, then resource utilization improves, but performance may deteriorate below threshold
Solution Approach 1:
The patent implements a feedback mechanism where the system evaluates the performance of compressed models against predetermined thresholds. When performance falls below the threshold, the system provides feedback to generate a dedicated model trained on specific data, ensuring that reliability requirements are met while still benefiting from compression where appropriate.
Solution Approach 2:
The patent creates a dynamic model selection system that adapts between compressed generic-purpose models and dedicated models based on performance requirements. This dynamic approach allows the system to use resource-efficient compressed models when performance thresholds are satisfied, and automatically switch to more resource-intensive dedicated models when higher reliability is needed.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The compressed artificial intelligence model reduces resource consumption while maintaining performance, enabling efficient operation and effective resource utilization in electronic apparatuses.
Implementation Method 1
acquire a rank of a singular value decomposition (SVD) algorithm based on a compression rate, compress and train the selected generic-purpose artificial intelligence model based on the acquired rank
Data Source
AI summary
A method for controlling an electronic apparatus is provided. The method for controlling an electronic apparatus includes the steps of selecting a generic-purpose artificial intelligence model, generating a compressed artificial intelligence model based on the selected generic-purpose artificial intelligence model, and generating a dedicated artificial intelligence model based on the generated compressed artificial intelligence model, and the step of generating a compressed artificial intelligence model includes the steps of acquiring a rank of a singular value decomposition (SVD) algorithm based on a compression rate, compressing and training the selected generic-purpose artificial intelligence model based on the acquired rank and converting the model into the compressed artificial intelligence model, determining the performance of the converted compressed artificial intelligence model based on a predetermined first threshold value, and based on the performance of the converted compressed artificial intelligence model being lower than the predetermined first threshold value, generating the dedicated artificial intelligence model.