Soft Augmentation
Soft augmentation is a data augmentation technique used to improve the robustness and generalization of machine learning models, particularly in scenarios with limited or noisy data. Current research focuses on developing application-specific augmentation strategies, often integrated with contrastive learning or teacher-student frameworks, across diverse domains including image processing, natural language processing, and time series analysis. These techniques aim to enhance model performance by increasing data diversity without significantly altering the underlying data distribution, leading to more reliable and generalizable models for various applications. The impact is seen in improved accuracy and robustness in tasks ranging from object detection and image dehazing to code-switched language understanding and medical image analysis.
Papers
Generative models improve fairness of medical classifiers under distribution shifts
Ira Ktena, Olivia Wiles, Isabela Albuquerque, Sylvestre-Alvise Rebuffi, Ryutaro Tanno, Abhijit Guha Roy, Shekoofeh Azizi, Danielle Belgrave, Pushmeet Kohli, Alan Karthikesalingam, Taylan Cemgil, Sven Gowal
Performance of GAN-based augmentation for deep learning COVID-19 image classification
Oleksandr Fedoruk, Konrad Klimaszewski, Aleksander Ogonowski, Rafał Możdżonek