Feature Request: add MixMatch and FixMatch training support for semi-supervised learning (SSL)

[surajyadav-research]

Why this feature

MixMatch and FixMatch are widely adopted SSL baselines and would make this repository more complete for modern semi-supervised workflows.

---

References

1. MixMatch: A Holistic Approach to Semi-Supervised Learning (NeurIPS 2019)

• Paper: https://papers.neurips.cc/paper/2019/file/1cd138d0499a68f4bb72bee04bbec2d7-Paper.pdf
• Citations: ~4571 (approx.)

2. FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020)

• Paper: https://papers.nips.cc/paper_files/paper/2020/file/06964dce9addb1c5cb5d6e3d9838f733-Paper.pdf
• Citations: ~5438 (approx.)

[aviralgarg05]

I would like to solve the issue. Please assign me the issue

[HarshRaj29004]

Hi @surajyadav-research I would like to help you with this
we could start with simple helper function like pseudo labeling and confidence masking, rather than a full training pipeline so users can plug them into their own loops

[surajyadav-research]

Hi @HarshRaj29004 @aviralgarg05 , thanks for the suggestion and for offering to help. This is already part of the review process, and I’d like to work on it myself for now. Happy to collaborate on a new or separate feature in the future.