Research map / Video / Video-language understanding
WatchLog: Efficient and Interpretable Event Reasoning for Endpoint Detection and Response Logs with Multimodal LLMs
Hongyi Zhou, Jianfeng Pan, Min Peng, Shaomang Huang, Xuling Zhang
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 image-event alignment pre-training (ablation) | WatchLog · EDR8M-20R | Overall accuracy: 5.3 -> 88.8 | Improved |
| add video-log alignment pre-training on top of image-event alignment (ablation) | WatchLog · EDR8M-20R | Overall accuracy: 88.8 -> 90.3 | Improved |
| add alignment pre-training (ablation) | WatchLog · EDR8M-20R | False alarm: 100.0 -> 5.0 | Improved |
| add image-event alignment pre-training (ablation) | WatchLog · EDR8M-20R | Rationale completeness: 32.1 -> 67.2 | Improved |
| add video-log alignment pre-training (ablation) | WatchLog · EDR8M-20R | Rationale completeness: 67.2 -> 68.5 | Improved |
| scale backbone from 2B to 3B (ablation) | WatchLog · EDR8M-20R | Overall accuracy: 90.3 -> 91.3 | Improved |
| scale backbone from 2B to 7B (ablation) | WatchLog · EDR8M-20R | Overall accuracy: 90.3 -> 92.5 | Improved |
| scale backbone from 2B to 7B (ablation) | WatchLog · EDR8M-20R | Rationale completeness: 68.5 -> 78.2 | Improved |
| scale backbone from 2B to 3B (ablation) | WatchLog · EDR8M-20R | False alarm: 5.0 -> 0.0 | Improved |
| replace long-context LLM with WatchLog video encoding (ablation) | WatchLog vs Qwen2.5-1.5B-Instruct · 1024K-token input | GPU memory: 98.46 -> 48.61 | Improved |
| replace long-context LLM with WatchLog video encoding (ablation) | WatchLog vs Qwen2.5-1.5B-Instruct · 1024K-token input | TTFT: 259.30 -> 1.11 | Improved |
| replace long-context LLM with WatchLog video encoding (ablation) | WatchLog vs Qwen2.5-7B-Instruct · 1024K-token input | TTFT: 486.19 -> 1.11 | Improved |
Closest papers by meaning
- A Unified Reasoning Framework for Holistic Zero-Shot Video Anomaly Analysis (NeurIPS 2025 · similarity 0.84)
- MoniTor: Exploiting Large Language Models with Instruction for Online Video Anomaly Detection (NeurIPS 2025 · similarity 0.84)
- OVO-Bench: How Far is Your Video-LLMs from Real-World Online Video Understanding? (CVPR 2025 · similarity 0.83)
- VADTree: Explainable Training-Free Video Anomaly Detection via Hierarchical Granularity-Aware Tree (NeurIPS 2025 · similarity 0.83)
- VERA: Explainable Video Anomaly Detection via Verbalized Learning of Vision-Language Models (CVPR 2025 · similarity 0.83)
- SLS-Bench: A Benchmark for Incident Log Summarization with Synthetic Observability Data (NeurIPS 2026 · similarity 0.82)
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.
