Similarity digest schemes provide an efficient way to measure similarity between arbitrary digital artifacts (e.g. files), by first mapping each input to a usually smaller sized similarity digest, and doing fast comparisons on these digests alone to determine a score on how similar or different the corresponding original inputs were.

Such schemes are mostly used in security related applications, like similarity based malware detection or filtering digital forensic artifacts to be further analyzed, basing the security of these applications on the requirement that an attacker cannot easily mislead the schemes applied. This desired adversarial robustness of most well known similarity digest schemes was challenged by quite a few proposed targeted attacks. However, in a lack of detailed experimental results and available implementation, it is hard to compare the true impact of such attacks, especially in a meaningful comparison regarding the implied adversarial robustness of the targeted schemes.

Tasks to be performed by the student will include:

• Implement some attacks proposed in literature against well known schemes, forming a toolkit of proposed adversarial techniques;
• Propose comparable metrics regarding how successful an attack is;
• Do measurements by executing said attacks and determining said metrics;
• Draw conclusions based on the comparative impact of the analyzed attacks on the adversarial robustness of the targeted schemes.