Sequential Data Interval Augmentation for Robust Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning methods require advanced preprocessing and interpolation processing to handle changes in data conditions, particularly in audio data, which complicates the generation of high-quality learning data.
Innovation Solution
A machine learning method that generates learning data by performing preprocessing for size adjustment on sequential data based on predetermined conditions, creating multiple pieces of adjusted sequential data with varying intervals, and performing supervised learning to develop a robust learning model without requiring advanced preprocessing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If advanced preprocessing and interpolation processing are performed on audio data, then the quality of learning data is improved, but the complexity of the processing system increases
Solution Approach 1:
The patent extracts and removes the complex advanced preprocessing and interpolation processing steps from the audio data handling pipeline. Instead of using complex processing to improve data quality, the invention directly uses the raw sequential data after simple preprocessing, thereby reducing system complexity while maintaining learning data quality through alternative means (data augmentation and interval adjustment).
Solution Approach 2:
The patent changes the approach from processing audio data through complex interpolation to adjusting the time intervals between sequential data points. By varying the sampling rate and creating multiple sequences with different intervals, the system achieves data diversity without requiring advanced preprocessing, thus improving data quality while simplifying the processing system.
2Quantity of substance
If the sampling rate is increased to generate more learning data, then the quantity of learning data is improved, but the robustness against condition changes may decrease
Solution Approach 1:
The patent applies dynamics by making the sampling rate adjustable and adaptive rather than fixed. The system can dynamically change the time intervals between sequential data points based on the specific conditions and requirements. This allows the system to generate sufficient learning data while maintaining robustness, as the sampling rate can be adjusted to match different operational conditions.
Solution Approach 2:
The patent segments the sequential data into multiple sequences with different time intervals. By dividing the original data stream into multiple subsets with varying sampling rates, the system creates diverse learning data that represents different conditions. This segmentation approach increases the quantity of learning data while improving robustness, as the model learns to handle various sampling conditions.
3Ease of operation
If simple preprocessing is used instead of advanced preprocessing, then the ease of operation is improved, but the quality of learning data may deteriorate
Solution Approach 1:
The patent uses copying and augmentation techniques to create multiple versions of the sequential data with different time intervals. Instead of relying on complex preprocessing to enhance data quality, the system creates augmented copies of the original data by adjusting sampling rates and time intervals. This copying approach maintains ease of operation while improving data quality through data augmentation rather than complex processing.
Data Source
AI summary
A machine learning method includes: acquiring sequential data; performing preprocessing for size adjustment in a sequential direction on the sequential data based on a predetermined condition to generate a plurality of pieces of adjusted sequential data having different intervals in the sequential direction from one piece of the sequential data; and performing supervised learning using the plurality of generated pieces of adjusted sequential data to generate a learning model.