TraceGuard: Adaptive Multimodal Poison Filtering through Cross-Feature Rank Agreement
cs.CR, cs.LG
Submitted: 2026-09-24
Updated: 2026-09-24
Terminology
Sources
- Making Laplacians commute
- Poisoning Web-Scale Training Datasets is Practical
- Detecting Backdoor Attacks on Deep Neural Networks by Activation Clustering
- CBV: Clean-label Backdoor Attacks on Vision Language Models via Diffusion Models
- Backdooring CLIP through Concept Confusion
- Distilling Cognitive Backdoor Patterns within an Image
- BackdoorIDS: Zero-shot Backdoor Detection for Pretrained Vision Encoder
- Understanding Neural Networks through Representation Erasure
- VL-Trojan: Multimodal Instruction Backdoor Attacks against Autoregressive Visual Language Models
- BadCLIP++: Stealthy and Persistent Backdoors in Multimodal Contrastive Learning
- TrojVLM: Backdoor Attack Against Vision Language Models
- Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs
- A Study of Backdoors in Instruction Fine-tuned Language Models
- LAION-400M: Open Dataset of CLIP-Filtered 400 Million Image-Text Pairs
- Concept-Guided Backdoor Attack on Vision Language Models
- Label-Consistent Backdoor Attacks
- Finetuned Multimodal Language Models Are High-Quality Image-Text Data Filters
- Robust Anti-Backdoor Instruction Tuning in LVLMs
- Adjustable Text-Guided Backdoor Attacks with Natural-Word Triggers on Multimodal Pretrained Models
- TokenSwap: Backdoor Attack on the Compositional Understanding of Large Vision-Language Models
Related papers
- SoK: AI-Augmented Binary Reversing
- Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
- Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
- Efficient Fuzzy PSI under One-Sided Assumptions
- Sealing the Audit-Runtime Gap for LLM Skills
- Token Composition: A Graph Based on EVM Logs