Reproduce “Debiasing, calibrating, and improving Semi-supervised Learning performance via simple Ensemble Projector” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Khanh-Binh Nguyen
Khanh-Binh Nguyen from Sungkyunkwan University introduced Ensemble Projectors Aided for Semi-supervised Learning (EPASS), a module designed to debias and calibrate contrastive joint-training semi-supervised learning models by replacing a single projector with an ensemble. This method consistently boosted classification performance, accelerated convergence, and improved model calibration across various datasets.
Key insights
- Relying on a single projector in contrastive joint-training SSL can lead to biased embeddings and propagate errors through the memory bank, limiting overall model performance.
- Employing an ensemble of projectors, each learning a distinct view, produces more generalized and less biased embeddings when aggregated.
- Averaging the outputs of three projectors offers an optimal balance between performance improvement and computational overhead for creating robust representations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.