Cascaded Test Time Augmentation for Corrupted Data Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data augmentation methods, such as single augmentation in test time data augmentation (TTA), are inadequate for severely corrupted test data, leading to suboptimal prediction results in machine learning models.
Innovation Solution
A method involving a cascaded test time augmentation (TTA) process using a trained first model to generate a target augmentation task sequence, which is then applied to the target data to generate augmented data, and subsequently processed by a second model to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If single augmentation is performed for each piece of test data, then the processing speed is maintained, but the prediction accuracy deteriorates for severely corrupted test data
Solution Approach 1:
The patent segments the augmentation process into multiple iterative steps, where each step generates and evaluates augmented samples separately. The model processes test data through multiple augmentation iterations, evaluating each augmented sample independently and selecting the best prediction, thereby improving accuracy for corrupted data while maintaining manageable processing throughput
Solution Approach 2:
The patent implements periodic action by performing data augmentation at multiple discrete time steps during inference. Instead of a single augmentation, the system periodically applies different augmentation operations across multiple iterations, allowing the model to adaptively select the most suitable augmented sample for prediction, thus improving accuracy without continuous processing
2Measurement precision
If multiple augmentation tasks are performed iteratively, then the prediction accuracy improves, but the computational cost increases
Solution Approach 1:
The patent applies partial action by performing a limited number of augmentation iterations (e.g., 3-5 iterations) rather than exhaustive search. This partial iteration approach achieves sufficient accuracy improvement for corrupted test data while avoiding the prohibitive computational cost of exhaustive multiple augmentations, balancing performance gain with computational efficiency
Solution Approach 2:
The patent performs preliminary action by pre-defining a set of candidate augmentation operations before the iterative process. These augmentation tasks are prepared in advance, and during inference, the system selectively applies them based on data characteristics, reducing the computational overhead of dynamically generating augmentation operations during each iteration
Data Source
AI summary
Disclosed is a method and device for processing data, and the method includes generating a target augmentation task sequence by processing the target data with a trained first model that performs inference on the target data to generate the target data augmentation task sequence, generate augmented target data by performing data augmentation on the target data according to the target augmentation task sequence, and obtaining a prediction result corresponding to the target data by inputting the augmented target data to a trained second model and performing a corresponding processing on the augmented target data by the trained second model.