Label-Free Localization and Test-Time Suppression of CLIP's Typographic Circuits
cs.CV
Submitted: 2026-11-01
Updated: 2026-11-01
Terminology
Sources
- Qwen3-VL Technical Report
- From What to How: Attributing CLIP's Latent Components Reveals Unexpected Semantic Reliance
- Interpreting the Second-Order Effects of Neurons in CLIP
- Towards Understanding the Nature of Attention with Low-Rank Sparse Decomposition
- Dyslexify: A Mechanistic Defense Against Typographic Attacks in CLIP
- Interpreting Attention Layer Outputs with Sparse Autoencoders
- Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders
- From Concepts to Components: Concept-Agnostic Attention Module Discovery in Transformers
- Gemma 3 Technical Report
- InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency
- Read or Ignore? A Unified Benchmark for Typographic-Attack Robustness and Text Recognition in Vision-Language Models
- Qwen-Image Technical Report
Related papers
- TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models
- Do Egocentric Video-Language Models Capture Both Hand- and Object-Centric Cues?
- Aura: Consistent Multi-Subject Video Generation via VLM-Grounded Semantic Alignment
- PVRF: All-in-one Adverse Weather Removal via Prior-modulated and Velocity-constrained Rectified Flow
- DLWM: Diverse Latent World Models for Efficient Multimodal Reasoning
- Out of Sight, Still in Mind: Token Compression for Omni-LLMs