Annotation Data Cleansing Using Model Feedback for Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning methods require manual cleansing of annotation data, which is time-consuming and prone to errors, especially when dealing with a large number of classes, leading to increased costs and limited accuracy.
Innovation Solution
An information processing system that automatically cleanses annotation data through a simpler method by generating a model using a data set, comparing classes, and excluding erroneous data, thereby improving the accuracy of deep learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual cleansing of annotation data is performed, then annotation accuracy can be improved, but processing time and labor costs increase significantly
Solution Approach 1:
The system uses the trained deep learning model to automatically identify and flag potentially erroneous annotation data. The model serves itself by using its own learned knowledge to detect inconsistencies in the training data, eliminating the need for manual review of all annotation samples while maintaining high accuracy standards.
Solution Approach 2:
Manual cleansing operations are replaced with automated computational processes. The deep learning model systematically analyzes annotation data using algorithmic criteria, substituting human experts with an automated system that can process large volumes of data quickly and consistently without fatigue or attention degradation.
2Measurement precision
If manual cleansing is performed for a large number of classes, then annotation accuracy can be maintained, but the complexity and cost of the process increase
Solution Approach 1:
The deep learning model performs multiple functions: it trains on the annotation data and simultaneously evaluates the quality of that same data. This universal approach works across any number of classes without requiring different methodologies, as the model adapts to the specific classification task while maintaining a consistent automated cleansing mechanism.
Solution Approach 2:
The system implements a feedback loop where the trained model's predictions are compared against the original annotations. Discrepancies between model predictions and annotated labels serve as feedback signals to identify potentially erroneous annotations, creating an automated quality control mechanism that scales to any number of classes without increasing process complexity.
3Measurement precision
If over-learning is performed to improve model accuracy, then classification performance may improve, but processing time and computational load increase
Solution Approach 1:
Instead of performing exhaustive over-learning with massive computational resources, the system applies a moderate training regime followed by automated quality assessment. The deep learning model is trained to sufficient convergence to learn meaningful patterns, then used to evaluate annotation quality, achieving adequate classification accuracy without the excessive computational burden of prolonged training.
Data Source
AI summary
An information processing system that executes cleansing processing of a data set that is a set of annotation data including classification target data and a class corresponding to the classification target data includes a learning processing unit that generates a model by executing learning processing of deep learning using the data set, and a cleansing processing unit that executes cleansing processing of annotation data in the data set. The cleansing processing unit executes the cleansing processing of first annotation data in a first data set by outputting a first class corresponding to classification target data in the first annotation data in the first data set and a second class by inputting the classification target data to a first model generated by the learning processing unit using the first data set, and comparing the first class with the second class.