Reproduce “IoT Device Labeling Using Large Language Models” 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

Bar Meyuhas, Anat Bremler-Barr, Tal Shapira

Researchers from Reichman University and Tel-Aviv University developed an automated, passive system for labeling previously unseen IoT devices by identifying their vendor and function. The system leverages large language models (LLMs) and zero-shot classification on enriched network traffic features, achieving 86% HIT1 accuracy for vendor identification and 70% HIT1 accuracy for function identification on a dataset of unique IoT devices.

Key insights

  • Enriching raw network features with information from search engine results significantly enhances both vendor and function labeling accuracy by providing crucial contextual data.
  • Vendor identification is most effectively handled by straightforward string matching, which is robust for a large number of vendor labels and less dependent on complex semantic understanding.
  • Zero-shot classification with LLMs is highly suitable for function identification due to its ability to semantically understand and classify device functions without explicit training on new categories, especially when guided by a vendor-restricted function catalog.

Summary by alphaXiv.

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