Skip to main content
Search

Khalifa University of Science and Technology

2 related articles

Subscribe to updates
  1. arXiv

    KaliBench's 8,504 query-command pairs show LLMs pick the right Kali tool but stumble on arguments

    Researchers built KaliBench, a fine-grained benchmark for schema-free natural-language-to-CLI cybersecurity tool use on Kali Linux, containing 8,504 query-command pairs verified by LLM checking, sandboxed execution, and human review across 1,642 tools, 23 capability dimensions, and five security phases; evaluating 24 open-weight model configurations under unrestricted, restricted, and hinted settings, they found tool selection rises from 72.0% to 95.2% on average once candidates are constrained while optional-argument F1 barely moves, and only hints lift optional-argument F1 from 45.1% to 87.8% and Exact Correct from 22.3% to 73.

    The University of Western AustraliaKhalifa University of Science and Technology
    Pengfei Li（First author）
    KaliBenchNatural-Language-to-CLI TranslationReinforcement Learning with Verifiable RewardsSupervised Fine-tuning
  2. IEEE Journal on Selected Areas in Communications

    FedFit cuts federated LoRA fine-tuning communication by up to 100x while matching standard federated LoRA perplexity via dual vector-bank parameterization and quantization

    FedFit targets the large communication overhead of federated fine-tuning of large language models and the LoRA aggregation dilemma between accurate Sum-of-Products (SoP) and communication-efficient Product-of-Sums (PoS); it reconstructs high-dimensional adapter matrices from two compact, disjoint global vector banks, alternates between decoupled single-bank updates and joint updates corrected by a Residual Spectral Aggregation mechanism, and adds blockwise quantization with client-side error feedback, achieving perplexity comparable to standard federated LoRA on Qwen2.5 models with compression ratios up to 100x higher, alongside theoretical convergence guarantees.

    Central South UniversityCentre National de la Recherche ScientifiqueUniversité de LorraineKhalifa University of Science and Technology
    Hang Zou（First author）
    Federated LoRA Fine-TuningPost-Training Quantization

You’re all caught up