Research map / Large language models / Language, safety and interpretability
HyperGLLM: An Efficient Framework for Endpoint Threat Detection via Hypergraph-Enhanced Large Language Models
Hongyi Zhou, Jianfeng Pan, Min Peng, Shaomang Huang, Hanzhong Zheng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add differential hypergraph network (DHGNN) (ablation) | HyperGLLM · EDR3.6B-63F | Binary accuracy: 93.90 -> 99.11 | Improved |
| add differential hypergraph network (DHGNN) (ablation) | HyperGLLM · EDR3.6B-63F | Overall accuracy: 88.12 -> 94.65 | Improved |
| add global differential mechanism (ablation) | HyperGLLM · EDR3.6B-63F | Binary accuracy: 94.94 -> 99.11 | Improved |
| add global differential mechanism (ablation) | HyperGLLM · EDR3.6B-63F | Overall accuracy: 89.58 -> 94.65 | Improved |
| use multi-granularity clustering instead of coarsest single level (ablation) | HyperGLLM · EDR3.6B-63F | Overall accuracy: 84.34 -> 94.65 | Improved |
| use multi-granularity clustering instead of finest single level (ablation) | HyperGLLM · EDR3.6B-63F | Overall accuracy: 85.29 -> 94.65 | Improved |
| raise max clustering scale S from 3 to 6 (ablation) | HyperGLLM · EDR3.6B-63F | Binary accuracy: 94.53 -> 99.11 | Improved |
| raise max clustering scale S from 6 to 7 (ablation) | HyperGLLM · EDR3.6B-63F | Binary accuracy: 99.11 -> 94.14 | Hurt |
| add hypergraph compression to the LLM versus fine-tuned long-context LLM | HyperGLLM vs LongRoPE · EDR3.6B-63F | Overall accuracy: 86.31 -> 94.65 | Improved |
| replace long-context LLM with HyperGLLM on unseen attacks | HyperGLLM vs LongRoPE · EDR3.6B-63F unseen attacks | Binary accuracy: 86.20 -> 97.98 | Improved |
| replace long-context LLM with HyperGLLM on ATLASv2 | HyperGLLM vs LongRoPE · ATLASv2 (host 2) | Binary accuracy: 69.4 -> 88.1 | Improved |
| use HyperGLLM instead of baseline LLMs at 1M-token input | HyperGLLM vs Qwen2.5-3B/LLaMA3-8B · EDR3.6B-63F | GPU memory and TTFT: less than 1/15 and 1/1000 of the baselines | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ScamNet — ScamNet: Toward Explainable Large Language Model-Based Fraudulent Shopping Website Detection (AAAI 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- When Agents Go Rogue: Activation-Based Detection of Malicious Behaviors in Multi-Agent Systems (ICML 2026 · similarity 0.84)
- BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models (NeurIPS 2025 · similarity 0.84)
- BadChain: Backdoor Chain-of-Thought Prompting for Large Language Models (ICLR 2024 · similarity 0.84)
- Local Sparsity Enables Unsupervised LLM Safety Detection (NeurIPS 2026 · similarity 0.83)
- LT-Defense: Searching-free Backdoor Defense via Exploiting the Long-tailed Effect (NeurIPS 2024 · similarity 0.83)
- Probe before You Talk: Towards Black-box Defense against Backdoor Unalignment for Large Language Models (ICLR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
