Examp. of Locality Sensitive Hashing & their Usage for Malware Classification
No ratings
Presented at hacktivity2018 2018 by
Several tools has been proposed for malware classification and similarity detection of binary malware samples, however none of them can solve all issues. In my presentation, I'll cover the problematics of Locality Sensitive Hashes and provide some experimental information about the comparison of different LSH algorithms. SSDEEPS's base algorithm, spamsum was originally designed for spam email detection. Although it discoveres some similarity between binaries, it basically needs large equal pieces of the byte code. This only happens rarely and can easily be altered. One of the contenders, TLSH (TrendMicro Locality Sensitive Hash) is a more stable similarity matching process. I'm going to present the results of the comparison on a smaller size samples set (~30k samples). Using LSHs is easy and doesn't require huge computational resources so after the process was deemed useful and effective it was extended to a large malware database of multiple hundreds of terabytes of samples. The experiments focus on ransomware sample classification, so I'm also going to present some details related to hunting for fresh unknown malware samples of known groups.