New adversarial attack targets online handwriting recognition systems

Researchers have developed a new method for creating adversarial attacks on online handwriting recognition systems. Unlike previous image-based attacks that add spatial perturbations, this new approach focuses on temporal editing, inserting or deleting points in the time series data. This method uses temporal salience, guided by gradient-based activation mapping, to identify critical time steps for modification. Experiments on datasets like Unipen and CASIA-OLHWDB show that this temporal editing attack is more effective in one-shot black-box scenarios and better preserves the visual integrity of the handwriting compared to traditional spatial attacks. AI

IMPACT This research highlights a new vulnerability in AI systems processing sequential data, potentially impacting the security and reliability of applications like digital signature verification.

RANK_REASON Research paper detailing a novel adversarial attack method for a specific AI application (online handwriting recognition). [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• CASIA-OLHWDB
• deep learning
• gradient-based activation mapping
• online handwriting recognition
• temporal editing
• temporal salience

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →