RoPAWS: Robust Semi-supervised Representation Learning from Uncurated Data
Authors:
- Sangwoo MoSMSangwoo Mo
- Mido AssranMAMido Assran
Preprint
Abstract
Semi-supervised learning aims to train a model using limited labels. State-of-the-art semi-supervised methods for image classification such as PAWS rely on self-supervised representations learned with large-scale unlabeled but curated data. However, PAWS is often less effective when using real-world unlabeled data that is uncurated, e.g., contains out-of-class data. We propose RoPAWS, a robust extension of PAWS that can work with real-world unlabeled data. We first reinterpret PAWS as a generative classifier that models densities using kernel density estimation. From this probabilistic perspective, we calibrate its prediction based on the densities of labeled and unlabeled data, which leads to a simple closed-form solution from the Bayes' rule. We demonstrate that RoPAWS significantly improves PAWS for uncurated Semi-iNat by +5.3% and curated ImageNet by +0.4%.Keywords
All Related Versions
This publication has 26 references indexed in Scilit:
- OpenCoS: Contrastive Semi-supervised Learning for Handling Open-Set Unlabeled Data2023
- Open-Set Semi-Supervised Object Detection2022
- Towards Realistic Semi-supervised Learning2022
- FeatMatch: Feature-Based Augmentation for Semi-supervised Learning2020
- Multi-task Curriculum Framework for Open-Set Semi-supervised Learning2020
- Momentum Contrast for Unsupervised Visual Representation Learning2020
- Self-Training With Noisy Student Improves ImageNet Classification2020
- Label Propagation for Deep Semi-Supervised Learning2019
- Deep Residual Learning for Image Recognition2016
- Semi-supervised kernel density estimation for video annotationComputer Vision and Image Understanding, 2009