sidbav / av-vs-evasive

Malware Detection and Evasion

CSCE 689 -> TAMU, 2022 | ML Based Cyberdefenses - Competition

Description -> Machine Learning based AV - Binary Classification

@sid-38's blog post about competition can be found here

@sidbav's blog post about competition can be found here

Results can be found here

Stages of the competition

i. Defense ii. Attack

-> Defenders Challenge Specifications:

Deliverable: Self-contained docker image with model querying via HTTP requests.

Goals:

 FPR: 1% | TPR: 95%

Constraints:

 Memory: 1 GB max RAM
 Response time: 5 seconds per sample
 Warning: Timeouts will be considered evasions.

=======================================================================================================

-> Attackers Challenge Specifications:

Deliverable: Evasive Malware Binaries.

Goals:

 Evade the most models possible.

Constraints:

Maximum file size: 5MB of appended data.
Evasive sample execution in sandbox must be equivlent to original sample.

Minimum score for grading: At least one sample must bypass at least one model.