Reproduce “Random CapsNet Forest Model for Imbalanced Malware Type Classification Task” 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

Aykut Çayır, Uğur Ünal, Hasan Dağ

A Random Capsule Network Forest combines capsule-based image classification with bootstrap aggregation for malware family recognition under severe class imbalance. Trained directly on grayscale representations of Malimg and BIG2015 samples, the ensemble reaches 98.72% and 99.56% accuracy, respectively, with gains over single CapsNet models and relatively compact individual networks.

Key insights

  • Ensembling capsule models provides modest but consistent improvements over individual CapsNets, with the larger gain appearing on BIG2015.
  • The forest reduces errors for minority malware families, including the small Simda category, although bootstrap sampling is not presented as explicit class balancing.
  • The approach avoids hand-crafted image descriptors, transfer learning, augmentation, and class-weighted loss; its individual models remain compact, but ensemble size increases total parameters and training cost.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.