MimeLens identifies binary file types from arbitrary fragments

Researchers have developed MimeLens, a new system designed to accurately identify the content type of binary data fragments, even when they lack headers or are sampled from arbitrary positions within a file. Unlike previous methods that require whole-file access, MimeLens utilizes BERT-style encoders pretrained on randomly sampled binary chunks. This approach significantly outperforms existing tools like Magika and libmagic on challenging datasets, including mid-stream network packets and random disk blocks, though it comes with a higher latency cost on CPUs. AI

IMPACT Enhances data analysis in security and forensics by enabling content-type detection on fragmented binary data.

RANK_REASON Academic paper introducing a new method for binary fragment classification. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• BERT
• Hugging Face
• Magika
• MimeLens

• paper
• other

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