Outline
- malware detector based on deep learning
- domain challengers for evasion
- append attack
- slack attack
1.Malware detector based on deep learning
Feature extraction in static malware classification, but feature engineering is challenging and time consuming because a list of feature has to constantly evolve to capture adaptive adversary.
- solution :end to end learning
training a robust classifier
- training malconv on a production-scale dataset(full)
- use published dataset anderson+
- use small (mini)
2.Domain challenges for evasion
- Q: can apply adversarial ML to binaries ?
- E: original PE sample + adversarial noise = evasive pe sample
- A:existing evasion