Iterative Neural Network Block With Parameter Changes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Artificial neural networks (ANNs) face a conflict between achieving high output quality and low power consumption on hardware platforms with limited resources, where changing parameters for iterative processing increases power demand and reduces flexibility, leading to poorer classification accuracy.
Innovation Solution
Implementing an iterative block within the ANN with a controlled change of a portion of its parameters between iterations, optimizing parameter changes to balance flexibility and power usage, and utilizing memristors for efficient and energy-effective storage and processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If parameters of iterative block are changed during iterations to improve flexibility and classification accuracy, then classification accuracy is improved, but power consumption increases and hardware resources are exceeded
Solution Approach 1:
The patent applies local quality by selectively changing only a portion of parameters in the iterative block during iterations, rather than changing all parameters. This localized parameter modification maintains classification accuracy while reducing power consumption and hardware resource requirements.
Solution Approach 2:
The patent implements partial action by changing only a subset of parameters during iterative processing. This partial parameter change achieves sufficient classification accuracy without the excessive power consumption and hardware resource usage that would result from changing all parameters.
2Device complexity
If iterative block is implemented multiple times to reduce hardware size, then hardware resources are reduced, but flexibility decreases and classification accuracy deteriorates
Solution Approach 1:
The patent applies dynamics by introducing controlled parameter changes during iterative block execution. This dynamic parameter adjustment maintains system flexibility and classification accuracy while the iterative structure itself reduces hardware size through resource reuse.
Solution Approach 2:
The patent directly applies parameter changes by modifying a portion of parameters during iterations. This enables the system to maintain flexibility and accuracy with reduced hardware by leveraging parameter variability across iterations rather than requiring dedicated hardware for each processing stage.
3Adaptability or versatility
If all parameters are changed during iterations to maximize flexibility, then flexibility is improved, but power consumption increases disproportionately
Solution Approach 1:
The patent applies local quality by selectively changing only a portion of parameters during iterations rather than all parameters. This localized approach maintains necessary flexibility for accurate classification while avoiding disproportionate power consumption associated with changing all parameters.
Solution Approach 2:
The patent implements partial action by changing only a subset of parameters during iterative processing. This partial parameter change achieves sufficient flexibility for classification tasks without the excessive power consumption that would result from changing all parameters.
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
This approach reduces the size and power requirements of ANNs while improving classification accuracy, achieving a balance between flexibility and energy efficiency, particularly beneficial for vehicle control systems with limited resources.
Implementation Method 1
utilizing memristors for efficient and energy-effective storage and processing
Data Source
AI summary
A method which processes inputs in a sequence of layers to form outputs. Within an artificial neural network (ANN), at least one iterative block including one or more layer(s) is established, which is to be implemented multiple times. A number J of iterations is established, for which this iterative block is at most to be implemented. An input of the iterative block is mapped by the iterative block onto an output. This output is again fed to the iterative block as input and again mapped by the iterative block onto a new output. Once the iterative block has been implemented J-times, the output supplied by the iterative block is fed as the input to a following layer or is provided as output of the ANN. A portion of the parameters, which characterize the behavior of the layers in the iterative block, is changed during the switch between the iterations.