Augmentation Free Self Supervised Learning

Augmentation-free self-supervised learning (SSL) aims to learn robust representations from unlabeled data without relying on artificial data augmentations, which can distort the underlying data characteristics or require careful, task-specific design. Current research focuses on developing alternative strategies, such as leveraging inherent data structures (e.g., through clustering or identifying semantically similar regions) or employing self-training methods with regularization techniques to generate pseudo-labels and refine model predictions. This approach addresses limitations of augmentation-based SSL, offering the potential for more reliable and generalizable models across diverse applications, including visual question answering, 3D point cloud understanding, and robotic control.

Papers

July 19, 2023

STRAPPER: Preference-based Reinforcement Learning via Self-training Augmentation and Peer Regularization
Yachen Kang, Li He, Jinxin Liu, Zifeng Zhuang, Donglin Wang
Reinforcement Learning Consistency Regularization Reward Learning Preference Based Reinforcement Learning Peer Learning Augmentation Free Self Supervised Learning

June 6, 2023

Q: How to Specialize Large Vision-Language Models to Data-Scarce VQA Tasks? A: Self-Train on Unlabeled Images!
Zaid Khan, Vijay Kumar BG, Samuel Schulter, Xiang Yu, Yun Fu, Manmohan Chandraker
Large Vision Language Model Visual Question Answering Unlabeled Image Knowledge Based Visual Question Answering Augmentation Free Self Supervised Learning

May 31, 2023

Augmentation-aware Self-supervised Learning with Conditioned Projector
Marcin Przewięźlikowski, Mateusz Pyla, Bartosz Zieliński, Bartłomiej Twardowski, Jacek Tabor, Marek Śmieja
Data Augmentation Self Supervised Robust Representation Augmentation Free Self Supervised Learning

October 6, 2022

Data Augmentation-free Unsupervised Learning for 3D Point Cloud Understanding
Guofeng Mei, Cristiano Saltori, Fabio Poiesi, Jian Zhang, Elisa Ricci, Nicu Sebe, Qiang Wu
Point Cloud Point Cloud Understanding 3D Object Classification Point Feature Augmentation Free Self Supervised Learning

February 16, 2022

Unified smoke and fire detection in an evolutionary framework with self-supervised progressive data augment
Hang Zhang, Su Yang, Hongyong Wang, zhongyan lu, helin sun
Detection Model Evolutionary Approach Fire Detection Smoke Segmentation Simultaneous Identification Augmentation Free Self Supervised Learning

December 5, 2021

Augmentation-Free Self-Supervised Learning on Graphs
Namkyeong Lee, Junseok Lee, Chanyoung Park
Self Supervised Learning Graph Drawing Self Supervised Method Contrastive Augmentation Augmentation Free Self Supervised Learning