LiteSSLHub hosts open-source research code for lightweight semi-supervised learning and text mining.
Our projects explore how compact student models can learn effectively from limited labeled data through distillation, co-training, peer collaboration, and self-improvement. The goal is to make semi-supervised NLP methods practical, reproducible, and accessible for researchers working with constrained annotation budgets.
- DisCo: Code for the EMNLP 2023 paper "DisCo: Co-training Distilled Student Models for Semi-supervised Text Mining."
- PSNET: PyTorch implementation of "Lightweight Contenders: Navigating Semi-Supervised Text Mining through Peer Collaboration and Self Transcendence."
- Semi-supervised text mining
- Lightweight student models
- Knowledge distillation
- Co-training and peer collaboration
- Efficient NLP under limited supervision
Start with DisCo for co-training distilled student models, or PSNET for peer-collaborative semi-supervised training workflows.
- Add paper links, BibTeX, and reproduction instructions to each repository.
- Add dataset preparation notes for DisCo and PSNET.
- Add environment files with tested Python and PyTorch versions.
- Add lightweight smoke-test scripts so users can verify installation quickly.
- Pin DisCo and PSNET on the organization overview.
Issues, reproduction notes, and research discussions are welcome.