Machine Learning Model Parameter Training via Intermediate Feature Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep machine learning models often fail when applied to data with noise not included in the training data, as they struggle to produce accurate output results due to insufficient training data coverage and incomplete data augmentation, leading to degraded learning accuracy.
Innovation Solution
A learning method that generates augmented training data to ensure specific intermediate feature values from the same original data are similar, using a back propagation method to learn parameters and increase the similarity of these feature values, thereby improving data separation and learning accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If data augmentation is applied to extend the range of training data, then the coverage of training data types is improved, but the similarity of intermediate feature values from the same original data deteriorates
Solution Approach 1:
The patent employs feedback by calculating the similarity of intermediate feature values from augmented data and using this information to adjust the augmentation process. The learning apparatus monitors whether augmented data maintains appropriate feature similarity and modifies the augmentation strategy accordingly, creating a closed-loop system that balances data diversity with feature consistency.
Solution Approach 2:
The patent changes parameters of the data augmentation process based on the calculated similarity of intermediate feature values. When similarity falls below a threshold, the augmentation parameters are adjusted to maintain feature consistency while still expanding data coverage. This dynamic parameter adjustment resolves the contradiction between extending data range and preserving feature similarity.
2Adaptability or versatility
If normalization is applied to put application data into the range of training data, then the compatibility between application data and training data is improved, but the handling complexity of data processing increases
Solution Approach 1:
The patent performs preliminary action by pre-processing training data to establish a standardized range and distribution before application data arrives. The learning apparatus prepares the training data framework in advance, defining the acceptable data range and feature distributions, so that when application data comes in, it can be more easily integrated without complex real-time normalization operations.
Data Source
AI summary
A learning method executed by a computer, the learning method includes: learning parameters of a machine learning model having intermediate feature values by inputting a plurality of augmented training data, which is generated by augmenting original training data, to the machine learning model so that specific intermediate feature values, which are calculated from specific augmented training data augmented from a same original training data, become similar to each other.