Reproduce “SemiScope: Disentangling Classifier Tuning and Joint Optimization in Semi-Supervised Security Classification” 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
Rui Shu, Tianpei Xia, Jingzhu He
SemiScope examines whether gains from optimizing semi-supervised security-classification pipelines come from coordinating SSL choices with the classifier or mainly from tuning the classifier itself. Across five imbalanced security datasets, joint optimization beats default SSL pipelines, but an equal-budget classifier-only search recovers most of those gains and is practically equivalent on four datasets under the study’s ±1-point margin.
Key insights
- Semi-supervised pipeline gains over defaults do not by themselves show that joint SSL optimization is responsible; a matched classifier-only search is needed to separate the effects.
- Classifier-family selection has the largest average impact in the ablation study, while confidence filtering has small and inconsistent effects across datasets.
- Validation thresholds matter for imbalanced classification: 99% of selected thresholds were below 0.5, with a mean of 0.20.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.