Researchers have developed a new method for creating adversarial attacks on deep learning models used for online handwriting recognition. Unlike existing spatial perturbation techniques that can introduce visible artifacts, this novel approach uses salience-guided temporal editing. By inserting or deleting points at critical time steps identified through gradient-based activation mapping, the method preserves the natural shape and smoothness of handwriting while generating adversarial examples. This temporal editing attack demonstrates stronger transferability in one-shot black-box scenarios compared to traditional image-based attacks, highlighting a significant threat model for handwriting recognition systems. AI
IMPACT This research highlights a new vulnerability in handwriting recognition AI, potentially impacting the security and reliability of systems relying on this technology.
RANK_REASON The cluster contains an academic paper detailing a new method for adversarial attacks on AI models.
- Adversarial Attacks and Defense Mechanisms to Improve Robustness of Deep Temporal Point Processes
- arXiv
- CASIA-OLHWDB
- deep learning
- gradient-based activation mapping
- online handwriting recognition
- temporal editing
- temporal salience
- Unipennate muscle
- alphaXiv
- CatalyzeX
- DagsHub
- Gotit.pub
- Hugging Face
- IArxiv
- one-shot black-box
- ScienceCast
- white box
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
