BackdoorVLM: A Benchmark for Backdoor Attacks and Defenses on Vision-Language Models
Backdoor attacks undermine the reliability and trustworthiness of machine learning systems by injecting hidden behaviors that can be maliciously activated at inference time. While such threats have been extensively studied in unimodal settings, their impact on multimodal foundation models, particularly vision-language
研究重點
Backdoor attacks undermine the reliability and trustworthiness of machine learning systems by injecting hidden behaviors that can be maliciously activated at inference time. While such threats have been extensively studied in unimodal settings, their impact on multimodal foundation models, particularly vision-language
為什麼值得關注
這個議題涉及 Crew Teacher、Crew Story 的實作方向。可先評估現有做法,再決定是否安排小型實驗。
可以進一步驗證什麼?
- 閱讀原文,確認它解決的問題、實驗方法及適用限制
- 檢查相關功能現有做法,找出可比較的差異
- 用小型 PoC 驗證成功率、品質、延遲和實際成本
參考資料
BackdoorVLM: A Benchmark for Backdoor Attacks and Defenses on Vision-Language Models ↗2026-11-05T00:00:00Z · 研究論文索引
延伸探索
相關研究
- Multimodal Resource-Exhaustion Attacks on Vision-Language Models via Joint Pixel-Prompt Optimization相同研究主題 · 2026-11-14
- When does online vision-language inference help? A survey of socially aware robot navigation相同研究主題 · 2026-11-12
- MetaNav: Metacognitive Vision-Language Navigation via Reflective Exploration相同研究主題 · 2026-11-08