Multimodal Resource-Exhaustion Attacks on Vision-Language Models via Joint Pixel-Prompt Optimization
Resource-exhaustion attacks against autoregressive vision-language models (VLMs) typically assume unimodal threat models, treating the image branch as the primary optimization surface while holding user-visible prompts fixed. Even recent loop-centric variants remain confined to this single-channel paradigm, leaving the
研究重點
Resource-exhaustion attacks against autoregressive vision-language models (VLMs) typically assume unimodal threat models, treating the image branch as the primary optimization surface while holding user-visible prompts fixed. Even recent loop-centric variants remain confined to this single-channel paradigm, leaving the
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
Multimodal Resource-Exhaustion Attacks on Vision-Language Models via Joint Pixel-Prompt Optimization ↗2026-11-12T00:00:00Z · 研究論文索引
延伸探索
相關研究
- BackdoorVLM: A Benchmark for Backdoor Attacks and Defenses on Vision-Language Models相同研究主題 · 2026-11-07
- Build specialized multimodal models with LFM3相同研究主題 · 2026-11-17
- When does online vision-language inference help? A survey of socially aware robot navigation相同研究主題 · 2026-11-12