Parallel Neural Network Channel Parameter Update via Dropout Error Difference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current parallel neural networks fail to improve classification precision due to uniform error distribution during backpropagation, which does not account for the varying strength of inputs from different data types such as images and time series, leading to ineffective learning and reduced classification accuracy.
Innovation Solution
A machine learning method that calculates output errors with and without dropout for each channel in a parallel neural network, identifying channels based on error differences to update parameters, allowing for targeted learning and improved classification precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If uniform backpropagation is used in parallel neural networks, then the network structure is simple and easy to implement, but classification precision deteriorates because error distribution does not account for varying input strengths
Solution Approach 1:
The patent applies local quality by differentiating error distribution across channels based on their specific characteristics. Instead of uniform error propagation, the system calculates channel-specific error distributions that account for varying input strengths and data types, allowing each channel to receive appropriately scaled error signals for optimal learning.
Solution Approach 2:
The patent changes the error distribution parameter from uniform to non-uniform across channels. By calculating and applying channel-specific error distribution parameters that reflect the actual importance and characteristics of each input channel, the system improves learning effectiveness without fundamentally changing the backpropagation framework.
2Adaptability or versatility
If features are extracted from the combination of plural inputs, then the network structure is simple, but it cannot handle different data sizes and types effectively
Solution Approach 1:
The patent segments the input processing into separate channels, where each channel handles a specific input type independently. This allows each channel to be optimized for its specific data characteristics (e.g., image data vs. time-series data) while maintaining overall network coherence through the parallel architecture and shared output layers.
Solution Approach 2:
The patent achieves universality by designing a parallel network architecture where multiple channels can process different data types through the same overall framework. The shared output layers and unified error calculation mechanism allow the system to handle diverse inputs (images, time-series, sensor data) without requiring completely separate processing systems.
Data Source
AI summary
A disclosed machine learning method includes: calculating a first output error between a label and an output in a case where dropout in which values are replaced with 0 is executed for a last layer of a first channel among plural channels in a parallel neural network; calculating a second output error between the label and an output in a case where the dropout is not executed for the last layer of the first channel; and identifying at least one channel from the plural channels based on a difference between the first output error and the second output error to update parameters of the identified channel.